xref: /openbmc/linux/net/packet/af_packet.c (revision aac5987a)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		PACKET - implements raw packet sockets.
7  *
8  * Authors:	Ross Biro
9  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
11  *
12  * Fixes:
13  *		Alan Cox	:	verify_area() now used correctly
14  *		Alan Cox	:	new skbuff lists, look ma no backlogs!
15  *		Alan Cox	:	tidied skbuff lists.
16  *		Alan Cox	:	Now uses generic datagram routines I
17  *					added. Also fixed the peek/read crash
18  *					from all old Linux datagram code.
19  *		Alan Cox	:	Uses the improved datagram code.
20  *		Alan Cox	:	Added NULL's for socket options.
21  *		Alan Cox	:	Re-commented the code.
22  *		Alan Cox	:	Use new kernel side addressing
23  *		Rob Janssen	:	Correct MTU usage.
24  *		Dave Platt	:	Counter leaks caused by incorrect
25  *					interrupt locking and some slightly
26  *					dubious gcc output. Can you read
27  *					compiler: it said _VOLATILE_
28  *	Richard Kooijman	:	Timestamp fixes.
29  *		Alan Cox	:	New buffers. Use sk->mac.raw.
30  *		Alan Cox	:	sendmsg/recvmsg support.
31  *		Alan Cox	:	Protocol setting support
32  *	Alexey Kuznetsov	:	Untied from IPv4 stack.
33  *	Cyrus Durgin		:	Fixed kerneld for kmod.
34  *	Michal Ostrowski        :       Module initialization cleanup.
35  *         Ulises Alonso        :       Frame number limit removal and
36  *                                      packet_set_ring memory leak.
37  *		Eric Biederman	:	Allow for > 8 byte hardware addresses.
38  *					The convention is that longer addresses
39  *					will simply extend the hardware address
40  *					byte arrays at the end of sockaddr_ll
41  *					and packet_mreq.
42  *		Johann Baudy	:	Added TX RING.
43  *		Chetan Loke	:	Implemented TPACKET_V3 block abstraction
44  *					layer.
45  *					Copyright (C) 2011, <lokec@ccs.neu.edu>
46  *
47  *
48  *		This program is free software; you can redistribute it and/or
49  *		modify it under the terms of the GNU General Public License
50  *		as published by the Free Software Foundation; either version
51  *		2 of the License, or (at your option) any later version.
52  *
53  */
54 
55 #include <linux/types.h>
56 #include <linux/mm.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
60 #include <linux/in.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
70 #include <net/ip.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
73 #include <net/sock.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <linux/uaccess.h>
77 #include <asm/ioctls.h>
78 #include <asm/page.h>
79 #include <asm/cacheflush.h>
80 #include <asm/io.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
92 #ifdef CONFIG_INET
93 #include <net/inet_common.h>
94 #endif
95 #include <linux/bpf.h>
96 #include <net/compat.h>
97 
98 #include "internal.h"
99 
100 /*
101    Assumptions:
102    - if device has no dev->hard_header routine, it adds and removes ll header
103      inside itself. In this case ll header is invisible outside of device,
104      but higher levels still should reserve dev->hard_header_len.
105      Some devices are enough clever to reallocate skb, when header
106      will not fit to reserved space (tunnel), another ones are silly
107      (PPP).
108    - packet socket receives packets with pulled ll header,
109      so that SOCK_RAW should push it back.
110 
111 On receive:
112 -----------
113 
114 Incoming, dev->hard_header!=NULL
115    mac_header -> ll header
116    data       -> data
117 
118 Outgoing, dev->hard_header!=NULL
119    mac_header -> ll header
120    data       -> ll header
121 
122 Incoming, dev->hard_header==NULL
123    mac_header -> UNKNOWN position. It is very likely, that it points to ll
124 		 header.  PPP makes it, that is wrong, because introduce
125 		 assymetry between rx and tx paths.
126    data       -> data
127 
128 Outgoing, dev->hard_header==NULL
129    mac_header -> data. ll header is still not built!
130    data       -> data
131 
132 Resume
133   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
134 
135 
136 On transmit:
137 ------------
138 
139 dev->hard_header != NULL
140    mac_header -> ll header
141    data       -> ll header
142 
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
144    mac_header -> data
145    data       -> data
146 
147    We should set nh.raw on output to correct posistion,
148    packet classifier depends on it.
149  */
150 
151 /* Private packet socket structures. */
152 
153 /* identical to struct packet_mreq except it has
154  * a longer address field.
155  */
156 struct packet_mreq_max {
157 	int		mr_ifindex;
158 	unsigned short	mr_type;
159 	unsigned short	mr_alen;
160 	unsigned char	mr_address[MAX_ADDR_LEN];
161 };
162 
163 union tpacket_uhdr {
164 	struct tpacket_hdr  *h1;
165 	struct tpacket2_hdr *h2;
166 	struct tpacket3_hdr *h3;
167 	void *raw;
168 };
169 
170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
171 		int closing, int tx_ring);
172 
173 #define V3_ALIGNMENT	(8)
174 
175 #define BLK_HDR_LEN	(ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
176 
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178 	(BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
179 
180 #define PGV_FROM_VMALLOC 1
181 
182 #define BLOCK_STATUS(x)	((x)->hdr.bh1.block_status)
183 #define BLOCK_NUM_PKTS(x)	((x)->hdr.bh1.num_pkts)
184 #define BLOCK_O2FP(x)		((x)->hdr.bh1.offset_to_first_pkt)
185 #define BLOCK_LEN(x)		((x)->hdr.bh1.blk_len)
186 #define BLOCK_SNUM(x)		((x)->hdr.bh1.seq_num)
187 #define BLOCK_O2PRIV(x)	((x)->offset_to_priv)
188 #define BLOCK_PRIV(x)		((void *)((char *)(x) + BLOCK_O2PRIV(x)))
189 
190 struct packet_sock;
191 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
192 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
193 		       struct packet_type *pt, struct net_device *orig_dev);
194 
195 static void *packet_previous_frame(struct packet_sock *po,
196 		struct packet_ring_buffer *rb,
197 		int status);
198 static void packet_increment_head(struct packet_ring_buffer *buff);
199 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
200 			struct tpacket_block_desc *);
201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
202 			struct packet_sock *);
203 static void prb_retire_current_block(struct tpacket_kbdq_core *,
204 		struct packet_sock *, unsigned int status);
205 static int prb_queue_frozen(struct tpacket_kbdq_core *);
206 static void prb_open_block(struct tpacket_kbdq_core *,
207 		struct tpacket_block_desc *);
208 static void prb_retire_rx_blk_timer_expired(unsigned long);
209 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
210 static void prb_init_blk_timer(struct packet_sock *,
211 		struct tpacket_kbdq_core *,
212 		void (*func) (unsigned long));
213 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
214 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
215 		struct tpacket3_hdr *);
216 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
217 		struct tpacket3_hdr *);
218 static void packet_flush_mclist(struct sock *sk);
219 
220 struct packet_skb_cb {
221 	union {
222 		struct sockaddr_pkt pkt;
223 		union {
224 			/* Trick: alias skb original length with
225 			 * ll.sll_family and ll.protocol in order
226 			 * to save room.
227 			 */
228 			unsigned int origlen;
229 			struct sockaddr_ll ll;
230 		};
231 	} sa;
232 };
233 
234 #define vio_le() virtio_legacy_is_little_endian()
235 
236 #define PACKET_SKB_CB(__skb)	((struct packet_skb_cb *)((__skb)->cb))
237 
238 #define GET_PBDQC_FROM_RB(x)	((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
239 #define GET_PBLOCK_DESC(x, bid)	\
240 	((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
241 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)	\
242 	((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
243 #define GET_NEXT_PRB_BLK_NUM(x) \
244 	(((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
245 	((x)->kactive_blk_num+1) : 0)
246 
247 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
248 static void __fanout_link(struct sock *sk, struct packet_sock *po);
249 
250 static int packet_direct_xmit(struct sk_buff *skb)
251 {
252 	struct net_device *dev = skb->dev;
253 	struct sk_buff *orig_skb = skb;
254 	struct netdev_queue *txq;
255 	int ret = NETDEV_TX_BUSY;
256 
257 	if (unlikely(!netif_running(dev) ||
258 		     !netif_carrier_ok(dev)))
259 		goto drop;
260 
261 	skb = validate_xmit_skb_list(skb, dev);
262 	if (skb != orig_skb)
263 		goto drop;
264 
265 	txq = skb_get_tx_queue(dev, skb);
266 
267 	local_bh_disable();
268 
269 	HARD_TX_LOCK(dev, txq, smp_processor_id());
270 	if (!netif_xmit_frozen_or_drv_stopped(txq))
271 		ret = netdev_start_xmit(skb, dev, txq, false);
272 	HARD_TX_UNLOCK(dev, txq);
273 
274 	local_bh_enable();
275 
276 	if (!dev_xmit_complete(ret))
277 		kfree_skb(skb);
278 
279 	return ret;
280 drop:
281 	atomic_long_inc(&dev->tx_dropped);
282 	kfree_skb_list(skb);
283 	return NET_XMIT_DROP;
284 }
285 
286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
287 {
288 	struct net_device *dev;
289 
290 	rcu_read_lock();
291 	dev = rcu_dereference(po->cached_dev);
292 	if (likely(dev))
293 		dev_hold(dev);
294 	rcu_read_unlock();
295 
296 	return dev;
297 }
298 
299 static void packet_cached_dev_assign(struct packet_sock *po,
300 				     struct net_device *dev)
301 {
302 	rcu_assign_pointer(po->cached_dev, dev);
303 }
304 
305 static void packet_cached_dev_reset(struct packet_sock *po)
306 {
307 	RCU_INIT_POINTER(po->cached_dev, NULL);
308 }
309 
310 static bool packet_use_direct_xmit(const struct packet_sock *po)
311 {
312 	return po->xmit == packet_direct_xmit;
313 }
314 
315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
316 {
317 	return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
318 }
319 
320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
321 {
322 	const struct net_device_ops *ops = dev->netdev_ops;
323 	u16 queue_index;
324 
325 	if (ops->ndo_select_queue) {
326 		queue_index = ops->ndo_select_queue(dev, skb, NULL,
327 						    __packet_pick_tx_queue);
328 		queue_index = netdev_cap_txqueue(dev, queue_index);
329 	} else {
330 		queue_index = __packet_pick_tx_queue(dev, skb);
331 	}
332 
333 	skb_set_queue_mapping(skb, queue_index);
334 }
335 
336 /* register_prot_hook must be invoked with the po->bind_lock held,
337  * or from a context in which asynchronous accesses to the packet
338  * socket is not possible (packet_create()).
339  */
340 static void register_prot_hook(struct sock *sk)
341 {
342 	struct packet_sock *po = pkt_sk(sk);
343 
344 	if (!po->running) {
345 		if (po->fanout)
346 			__fanout_link(sk, po);
347 		else
348 			dev_add_pack(&po->prot_hook);
349 
350 		sock_hold(sk);
351 		po->running = 1;
352 	}
353 }
354 
355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
356  * held.   If the sync parameter is true, we will temporarily drop
357  * the po->bind_lock and do a synchronize_net to make sure no
358  * asynchronous packet processing paths still refer to the elements
359  * of po->prot_hook.  If the sync parameter is false, it is the
360  * callers responsibility to take care of this.
361  */
362 static void __unregister_prot_hook(struct sock *sk, bool sync)
363 {
364 	struct packet_sock *po = pkt_sk(sk);
365 
366 	po->running = 0;
367 
368 	if (po->fanout)
369 		__fanout_unlink(sk, po);
370 	else
371 		__dev_remove_pack(&po->prot_hook);
372 
373 	__sock_put(sk);
374 
375 	if (sync) {
376 		spin_unlock(&po->bind_lock);
377 		synchronize_net();
378 		spin_lock(&po->bind_lock);
379 	}
380 }
381 
382 static void unregister_prot_hook(struct sock *sk, bool sync)
383 {
384 	struct packet_sock *po = pkt_sk(sk);
385 
386 	if (po->running)
387 		__unregister_prot_hook(sk, sync);
388 }
389 
390 static inline struct page * __pure pgv_to_page(void *addr)
391 {
392 	if (is_vmalloc_addr(addr))
393 		return vmalloc_to_page(addr);
394 	return virt_to_page(addr);
395 }
396 
397 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
398 {
399 	union tpacket_uhdr h;
400 
401 	h.raw = frame;
402 	switch (po->tp_version) {
403 	case TPACKET_V1:
404 		h.h1->tp_status = status;
405 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
406 		break;
407 	case TPACKET_V2:
408 		h.h2->tp_status = status;
409 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
410 		break;
411 	case TPACKET_V3:
412 		h.h3->tp_status = status;
413 		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
414 		break;
415 	default:
416 		WARN(1, "TPACKET version not supported.\n");
417 		BUG();
418 	}
419 
420 	smp_wmb();
421 }
422 
423 static int __packet_get_status(struct packet_sock *po, void *frame)
424 {
425 	union tpacket_uhdr h;
426 
427 	smp_rmb();
428 
429 	h.raw = frame;
430 	switch (po->tp_version) {
431 	case TPACKET_V1:
432 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
433 		return h.h1->tp_status;
434 	case TPACKET_V2:
435 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
436 		return h.h2->tp_status;
437 	case TPACKET_V3:
438 		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
439 		return h.h3->tp_status;
440 	default:
441 		WARN(1, "TPACKET version not supported.\n");
442 		BUG();
443 		return 0;
444 	}
445 }
446 
447 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
448 				   unsigned int flags)
449 {
450 	struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
451 
452 	if (shhwtstamps &&
453 	    (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
454 	    ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
455 		return TP_STATUS_TS_RAW_HARDWARE;
456 
457 	if (ktime_to_timespec_cond(skb->tstamp, ts))
458 		return TP_STATUS_TS_SOFTWARE;
459 
460 	return 0;
461 }
462 
463 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
464 				    struct sk_buff *skb)
465 {
466 	union tpacket_uhdr h;
467 	struct timespec ts;
468 	__u32 ts_status;
469 
470 	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
471 		return 0;
472 
473 	h.raw = frame;
474 	switch (po->tp_version) {
475 	case TPACKET_V1:
476 		h.h1->tp_sec = ts.tv_sec;
477 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
478 		break;
479 	case TPACKET_V2:
480 		h.h2->tp_sec = ts.tv_sec;
481 		h.h2->tp_nsec = ts.tv_nsec;
482 		break;
483 	case TPACKET_V3:
484 		h.h3->tp_sec = ts.tv_sec;
485 		h.h3->tp_nsec = ts.tv_nsec;
486 		break;
487 	default:
488 		WARN(1, "TPACKET version not supported.\n");
489 		BUG();
490 	}
491 
492 	/* one flush is safe, as both fields always lie on the same cacheline */
493 	flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
494 	smp_wmb();
495 
496 	return ts_status;
497 }
498 
499 static void *packet_lookup_frame(struct packet_sock *po,
500 		struct packet_ring_buffer *rb,
501 		unsigned int position,
502 		int status)
503 {
504 	unsigned int pg_vec_pos, frame_offset;
505 	union tpacket_uhdr h;
506 
507 	pg_vec_pos = position / rb->frames_per_block;
508 	frame_offset = position % rb->frames_per_block;
509 
510 	h.raw = rb->pg_vec[pg_vec_pos].buffer +
511 		(frame_offset * rb->frame_size);
512 
513 	if (status != __packet_get_status(po, h.raw))
514 		return NULL;
515 
516 	return h.raw;
517 }
518 
519 static void *packet_current_frame(struct packet_sock *po,
520 		struct packet_ring_buffer *rb,
521 		int status)
522 {
523 	return packet_lookup_frame(po, rb, rb->head, status);
524 }
525 
526 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
527 {
528 	del_timer_sync(&pkc->retire_blk_timer);
529 }
530 
531 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
532 		struct sk_buff_head *rb_queue)
533 {
534 	struct tpacket_kbdq_core *pkc;
535 
536 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
537 
538 	spin_lock_bh(&rb_queue->lock);
539 	pkc->delete_blk_timer = 1;
540 	spin_unlock_bh(&rb_queue->lock);
541 
542 	prb_del_retire_blk_timer(pkc);
543 }
544 
545 static void prb_init_blk_timer(struct packet_sock *po,
546 		struct tpacket_kbdq_core *pkc,
547 		void (*func) (unsigned long))
548 {
549 	init_timer(&pkc->retire_blk_timer);
550 	pkc->retire_blk_timer.data = (long)po;
551 	pkc->retire_blk_timer.function = func;
552 	pkc->retire_blk_timer.expires = jiffies;
553 }
554 
555 static void prb_setup_retire_blk_timer(struct packet_sock *po)
556 {
557 	struct tpacket_kbdq_core *pkc;
558 
559 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
560 	prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
561 }
562 
563 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
564 				int blk_size_in_bytes)
565 {
566 	struct net_device *dev;
567 	unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
568 	struct ethtool_link_ksettings ecmd;
569 	int err;
570 
571 	rtnl_lock();
572 	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
573 	if (unlikely(!dev)) {
574 		rtnl_unlock();
575 		return DEFAULT_PRB_RETIRE_TOV;
576 	}
577 	err = __ethtool_get_link_ksettings(dev, &ecmd);
578 	rtnl_unlock();
579 	if (!err) {
580 		/*
581 		 * If the link speed is so slow you don't really
582 		 * need to worry about perf anyways
583 		 */
584 		if (ecmd.base.speed < SPEED_1000 ||
585 		    ecmd.base.speed == SPEED_UNKNOWN) {
586 			return DEFAULT_PRB_RETIRE_TOV;
587 		} else {
588 			msec = 1;
589 			div = ecmd.base.speed / 1000;
590 		}
591 	}
592 
593 	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
594 
595 	if (div)
596 		mbits /= div;
597 
598 	tmo = mbits * msec;
599 
600 	if (div)
601 		return tmo+1;
602 	return tmo;
603 }
604 
605 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
606 			union tpacket_req_u *req_u)
607 {
608 	p1->feature_req_word = req_u->req3.tp_feature_req_word;
609 }
610 
611 static void init_prb_bdqc(struct packet_sock *po,
612 			struct packet_ring_buffer *rb,
613 			struct pgv *pg_vec,
614 			union tpacket_req_u *req_u)
615 {
616 	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
617 	struct tpacket_block_desc *pbd;
618 
619 	memset(p1, 0x0, sizeof(*p1));
620 
621 	p1->knxt_seq_num = 1;
622 	p1->pkbdq = pg_vec;
623 	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
624 	p1->pkblk_start	= pg_vec[0].buffer;
625 	p1->kblk_size = req_u->req3.tp_block_size;
626 	p1->knum_blocks	= req_u->req3.tp_block_nr;
627 	p1->hdrlen = po->tp_hdrlen;
628 	p1->version = po->tp_version;
629 	p1->last_kactive_blk_num = 0;
630 	po->stats.stats3.tp_freeze_q_cnt = 0;
631 	if (req_u->req3.tp_retire_blk_tov)
632 		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
633 	else
634 		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
635 						req_u->req3.tp_block_size);
636 	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
637 	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
638 
639 	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
640 	prb_init_ft_ops(p1, req_u);
641 	prb_setup_retire_blk_timer(po);
642 	prb_open_block(p1, pbd);
643 }
644 
645 /*  Do NOT update the last_blk_num first.
646  *  Assumes sk_buff_head lock is held.
647  */
648 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
649 {
650 	mod_timer(&pkc->retire_blk_timer,
651 			jiffies + pkc->tov_in_jiffies);
652 	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
653 }
654 
655 /*
656  * Timer logic:
657  * 1) We refresh the timer only when we open a block.
658  *    By doing this we don't waste cycles refreshing the timer
659  *	  on packet-by-packet basis.
660  *
661  * With a 1MB block-size, on a 1Gbps line, it will take
662  * i) ~8 ms to fill a block + ii) memcpy etc.
663  * In this cut we are not accounting for the memcpy time.
664  *
665  * So, if the user sets the 'tmo' to 10ms then the timer
666  * will never fire while the block is still getting filled
667  * (which is what we want). However, the user could choose
668  * to close a block early and that's fine.
669  *
670  * But when the timer does fire, we check whether or not to refresh it.
671  * Since the tmo granularity is in msecs, it is not too expensive
672  * to refresh the timer, lets say every '8' msecs.
673  * Either the user can set the 'tmo' or we can derive it based on
674  * a) line-speed and b) block-size.
675  * prb_calc_retire_blk_tmo() calculates the tmo.
676  *
677  */
678 static void prb_retire_rx_blk_timer_expired(unsigned long data)
679 {
680 	struct packet_sock *po = (struct packet_sock *)data;
681 	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
682 	unsigned int frozen;
683 	struct tpacket_block_desc *pbd;
684 
685 	spin_lock(&po->sk.sk_receive_queue.lock);
686 
687 	frozen = prb_queue_frozen(pkc);
688 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
689 
690 	if (unlikely(pkc->delete_blk_timer))
691 		goto out;
692 
693 	/* We only need to plug the race when the block is partially filled.
694 	 * tpacket_rcv:
695 	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
696 	 *		copy_bits() is in progress ...
697 	 *		timer fires on other cpu:
698 	 *		we can't retire the current block because copy_bits
699 	 *		is in progress.
700 	 *
701 	 */
702 	if (BLOCK_NUM_PKTS(pbd)) {
703 		while (atomic_read(&pkc->blk_fill_in_prog)) {
704 			/* Waiting for skb_copy_bits to finish... */
705 			cpu_relax();
706 		}
707 	}
708 
709 	if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
710 		if (!frozen) {
711 			if (!BLOCK_NUM_PKTS(pbd)) {
712 				/* An empty block. Just refresh the timer. */
713 				goto refresh_timer;
714 			}
715 			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
716 			if (!prb_dispatch_next_block(pkc, po))
717 				goto refresh_timer;
718 			else
719 				goto out;
720 		} else {
721 			/* Case 1. Queue was frozen because user-space was
722 			 *	   lagging behind.
723 			 */
724 			if (prb_curr_blk_in_use(pkc, pbd)) {
725 				/*
726 				 * Ok, user-space is still behind.
727 				 * So just refresh the timer.
728 				 */
729 				goto refresh_timer;
730 			} else {
731 			       /* Case 2. queue was frozen,user-space caught up,
732 				* now the link went idle && the timer fired.
733 				* We don't have a block to close.So we open this
734 				* block and restart the timer.
735 				* opening a block thaws the queue,restarts timer
736 				* Thawing/timer-refresh is a side effect.
737 				*/
738 				prb_open_block(pkc, pbd);
739 				goto out;
740 			}
741 		}
742 	}
743 
744 refresh_timer:
745 	_prb_refresh_rx_retire_blk_timer(pkc);
746 
747 out:
748 	spin_unlock(&po->sk.sk_receive_queue.lock);
749 }
750 
751 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
752 		struct tpacket_block_desc *pbd1, __u32 status)
753 {
754 	/* Flush everything minus the block header */
755 
756 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
757 	u8 *start, *end;
758 
759 	start = (u8 *)pbd1;
760 
761 	/* Skip the block header(we know header WILL fit in 4K) */
762 	start += PAGE_SIZE;
763 
764 	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
765 	for (; start < end; start += PAGE_SIZE)
766 		flush_dcache_page(pgv_to_page(start));
767 
768 	smp_wmb();
769 #endif
770 
771 	/* Now update the block status. */
772 
773 	BLOCK_STATUS(pbd1) = status;
774 
775 	/* Flush the block header */
776 
777 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
778 	start = (u8 *)pbd1;
779 	flush_dcache_page(pgv_to_page(start));
780 
781 	smp_wmb();
782 #endif
783 }
784 
785 /*
786  * Side effect:
787  *
788  * 1) flush the block
789  * 2) Increment active_blk_num
790  *
791  * Note:We DONT refresh the timer on purpose.
792  *	Because almost always the next block will be opened.
793  */
794 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
795 		struct tpacket_block_desc *pbd1,
796 		struct packet_sock *po, unsigned int stat)
797 {
798 	__u32 status = TP_STATUS_USER | stat;
799 
800 	struct tpacket3_hdr *last_pkt;
801 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
802 	struct sock *sk = &po->sk;
803 
804 	if (po->stats.stats3.tp_drops)
805 		status |= TP_STATUS_LOSING;
806 
807 	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
808 	last_pkt->tp_next_offset = 0;
809 
810 	/* Get the ts of the last pkt */
811 	if (BLOCK_NUM_PKTS(pbd1)) {
812 		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
813 		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
814 	} else {
815 		/* Ok, we tmo'd - so get the current time.
816 		 *
817 		 * It shouldn't really happen as we don't close empty
818 		 * blocks. See prb_retire_rx_blk_timer_expired().
819 		 */
820 		struct timespec ts;
821 		getnstimeofday(&ts);
822 		h1->ts_last_pkt.ts_sec = ts.tv_sec;
823 		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
824 	}
825 
826 	smp_wmb();
827 
828 	/* Flush the block */
829 	prb_flush_block(pkc1, pbd1, status);
830 
831 	sk->sk_data_ready(sk);
832 
833 	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
834 }
835 
836 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
837 {
838 	pkc->reset_pending_on_curr_blk = 0;
839 }
840 
841 /*
842  * Side effect of opening a block:
843  *
844  * 1) prb_queue is thawed.
845  * 2) retire_blk_timer is refreshed.
846  *
847  */
848 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
849 	struct tpacket_block_desc *pbd1)
850 {
851 	struct timespec ts;
852 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
853 
854 	smp_rmb();
855 
856 	/* We could have just memset this but we will lose the
857 	 * flexibility of making the priv area sticky
858 	 */
859 
860 	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
861 	BLOCK_NUM_PKTS(pbd1) = 0;
862 	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
863 
864 	getnstimeofday(&ts);
865 
866 	h1->ts_first_pkt.ts_sec = ts.tv_sec;
867 	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
868 
869 	pkc1->pkblk_start = (char *)pbd1;
870 	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
871 
872 	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
873 	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
874 
875 	pbd1->version = pkc1->version;
876 	pkc1->prev = pkc1->nxt_offset;
877 	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
878 
879 	prb_thaw_queue(pkc1);
880 	_prb_refresh_rx_retire_blk_timer(pkc1);
881 
882 	smp_wmb();
883 }
884 
885 /*
886  * Queue freeze logic:
887  * 1) Assume tp_block_nr = 8 blocks.
888  * 2) At time 't0', user opens Rx ring.
889  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
890  * 4) user-space is either sleeping or processing block '0'.
891  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
892  *    it will close block-7,loop around and try to fill block '0'.
893  *    call-flow:
894  *    __packet_lookup_frame_in_block
895  *      prb_retire_current_block()
896  *      prb_dispatch_next_block()
897  *        |->(BLOCK_STATUS == USER) evaluates to true
898  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
899  * 6) Now there are two cases:
900  *    6.1) Link goes idle right after the queue is frozen.
901  *         But remember, the last open_block() refreshed the timer.
902  *         When this timer expires,it will refresh itself so that we can
903  *         re-open block-0 in near future.
904  *    6.2) Link is busy and keeps on receiving packets. This is a simple
905  *         case and __packet_lookup_frame_in_block will check if block-0
906  *         is free and can now be re-used.
907  */
908 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
909 				  struct packet_sock *po)
910 {
911 	pkc->reset_pending_on_curr_blk = 1;
912 	po->stats.stats3.tp_freeze_q_cnt++;
913 }
914 
915 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
916 
917 /*
918  * If the next block is free then we will dispatch it
919  * and return a good offset.
920  * Else, we will freeze the queue.
921  * So, caller must check the return value.
922  */
923 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
924 		struct packet_sock *po)
925 {
926 	struct tpacket_block_desc *pbd;
927 
928 	smp_rmb();
929 
930 	/* 1. Get current block num */
931 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
932 
933 	/* 2. If this block is currently in_use then freeze the queue */
934 	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
935 		prb_freeze_queue(pkc, po);
936 		return NULL;
937 	}
938 
939 	/*
940 	 * 3.
941 	 * open this block and return the offset where the first packet
942 	 * needs to get stored.
943 	 */
944 	prb_open_block(pkc, pbd);
945 	return (void *)pkc->nxt_offset;
946 }
947 
948 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
949 		struct packet_sock *po, unsigned int status)
950 {
951 	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
952 
953 	/* retire/close the current block */
954 	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
955 		/*
956 		 * Plug the case where copy_bits() is in progress on
957 		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
958 		 * have space to copy the pkt in the current block and
959 		 * called prb_retire_current_block()
960 		 *
961 		 * We don't need to worry about the TMO case because
962 		 * the timer-handler already handled this case.
963 		 */
964 		if (!(status & TP_STATUS_BLK_TMO)) {
965 			while (atomic_read(&pkc->blk_fill_in_prog)) {
966 				/* Waiting for skb_copy_bits to finish... */
967 				cpu_relax();
968 			}
969 		}
970 		prb_close_block(pkc, pbd, po, status);
971 		return;
972 	}
973 }
974 
975 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
976 				      struct tpacket_block_desc *pbd)
977 {
978 	return TP_STATUS_USER & BLOCK_STATUS(pbd);
979 }
980 
981 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
982 {
983 	return pkc->reset_pending_on_curr_blk;
984 }
985 
986 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
987 {
988 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
989 	atomic_dec(&pkc->blk_fill_in_prog);
990 }
991 
992 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
993 			struct tpacket3_hdr *ppd)
994 {
995 	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
996 }
997 
998 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
999 			struct tpacket3_hdr *ppd)
1000 {
1001 	ppd->hv1.tp_rxhash = 0;
1002 }
1003 
1004 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1005 			struct tpacket3_hdr *ppd)
1006 {
1007 	if (skb_vlan_tag_present(pkc->skb)) {
1008 		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1009 		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1010 		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1011 	} else {
1012 		ppd->hv1.tp_vlan_tci = 0;
1013 		ppd->hv1.tp_vlan_tpid = 0;
1014 		ppd->tp_status = TP_STATUS_AVAILABLE;
1015 	}
1016 }
1017 
1018 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1019 			struct tpacket3_hdr *ppd)
1020 {
1021 	ppd->hv1.tp_padding = 0;
1022 	prb_fill_vlan_info(pkc, ppd);
1023 
1024 	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1025 		prb_fill_rxhash(pkc, ppd);
1026 	else
1027 		prb_clear_rxhash(pkc, ppd);
1028 }
1029 
1030 static void prb_fill_curr_block(char *curr,
1031 				struct tpacket_kbdq_core *pkc,
1032 				struct tpacket_block_desc *pbd,
1033 				unsigned int len)
1034 {
1035 	struct tpacket3_hdr *ppd;
1036 
1037 	ppd  = (struct tpacket3_hdr *)curr;
1038 	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1039 	pkc->prev = curr;
1040 	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1041 	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1042 	BLOCK_NUM_PKTS(pbd) += 1;
1043 	atomic_inc(&pkc->blk_fill_in_prog);
1044 	prb_run_all_ft_ops(pkc, ppd);
1045 }
1046 
1047 /* Assumes caller has the sk->rx_queue.lock */
1048 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1049 					    struct sk_buff *skb,
1050 						int status,
1051 					    unsigned int len
1052 					    )
1053 {
1054 	struct tpacket_kbdq_core *pkc;
1055 	struct tpacket_block_desc *pbd;
1056 	char *curr, *end;
1057 
1058 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1059 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1060 
1061 	/* Queue is frozen when user space is lagging behind */
1062 	if (prb_queue_frozen(pkc)) {
1063 		/*
1064 		 * Check if that last block which caused the queue to freeze,
1065 		 * is still in_use by user-space.
1066 		 */
1067 		if (prb_curr_blk_in_use(pkc, pbd)) {
1068 			/* Can't record this packet */
1069 			return NULL;
1070 		} else {
1071 			/*
1072 			 * Ok, the block was released by user-space.
1073 			 * Now let's open that block.
1074 			 * opening a block also thaws the queue.
1075 			 * Thawing is a side effect.
1076 			 */
1077 			prb_open_block(pkc, pbd);
1078 		}
1079 	}
1080 
1081 	smp_mb();
1082 	curr = pkc->nxt_offset;
1083 	pkc->skb = skb;
1084 	end = (char *)pbd + pkc->kblk_size;
1085 
1086 	/* first try the current block */
1087 	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1088 		prb_fill_curr_block(curr, pkc, pbd, len);
1089 		return (void *)curr;
1090 	}
1091 
1092 	/* Ok, close the current block */
1093 	prb_retire_current_block(pkc, po, 0);
1094 
1095 	/* Now, try to dispatch the next block */
1096 	curr = (char *)prb_dispatch_next_block(pkc, po);
1097 	if (curr) {
1098 		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1099 		prb_fill_curr_block(curr, pkc, pbd, len);
1100 		return (void *)curr;
1101 	}
1102 
1103 	/*
1104 	 * No free blocks are available.user_space hasn't caught up yet.
1105 	 * Queue was just frozen and now this packet will get dropped.
1106 	 */
1107 	return NULL;
1108 }
1109 
1110 static void *packet_current_rx_frame(struct packet_sock *po,
1111 					    struct sk_buff *skb,
1112 					    int status, unsigned int len)
1113 {
1114 	char *curr = NULL;
1115 	switch (po->tp_version) {
1116 	case TPACKET_V1:
1117 	case TPACKET_V2:
1118 		curr = packet_lookup_frame(po, &po->rx_ring,
1119 					po->rx_ring.head, status);
1120 		return curr;
1121 	case TPACKET_V3:
1122 		return __packet_lookup_frame_in_block(po, skb, status, len);
1123 	default:
1124 		WARN(1, "TPACKET version not supported\n");
1125 		BUG();
1126 		return NULL;
1127 	}
1128 }
1129 
1130 static void *prb_lookup_block(struct packet_sock *po,
1131 				     struct packet_ring_buffer *rb,
1132 				     unsigned int idx,
1133 				     int status)
1134 {
1135 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1136 	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1137 
1138 	if (status != BLOCK_STATUS(pbd))
1139 		return NULL;
1140 	return pbd;
1141 }
1142 
1143 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1144 {
1145 	unsigned int prev;
1146 	if (rb->prb_bdqc.kactive_blk_num)
1147 		prev = rb->prb_bdqc.kactive_blk_num-1;
1148 	else
1149 		prev = rb->prb_bdqc.knum_blocks-1;
1150 	return prev;
1151 }
1152 
1153 /* Assumes caller has held the rx_queue.lock */
1154 static void *__prb_previous_block(struct packet_sock *po,
1155 					 struct packet_ring_buffer *rb,
1156 					 int status)
1157 {
1158 	unsigned int previous = prb_previous_blk_num(rb);
1159 	return prb_lookup_block(po, rb, previous, status);
1160 }
1161 
1162 static void *packet_previous_rx_frame(struct packet_sock *po,
1163 					     struct packet_ring_buffer *rb,
1164 					     int status)
1165 {
1166 	if (po->tp_version <= TPACKET_V2)
1167 		return packet_previous_frame(po, rb, status);
1168 
1169 	return __prb_previous_block(po, rb, status);
1170 }
1171 
1172 static void packet_increment_rx_head(struct packet_sock *po,
1173 					    struct packet_ring_buffer *rb)
1174 {
1175 	switch (po->tp_version) {
1176 	case TPACKET_V1:
1177 	case TPACKET_V2:
1178 		return packet_increment_head(rb);
1179 	case TPACKET_V3:
1180 	default:
1181 		WARN(1, "TPACKET version not supported.\n");
1182 		BUG();
1183 		return;
1184 	}
1185 }
1186 
1187 static void *packet_previous_frame(struct packet_sock *po,
1188 		struct packet_ring_buffer *rb,
1189 		int status)
1190 {
1191 	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1192 	return packet_lookup_frame(po, rb, previous, status);
1193 }
1194 
1195 static void packet_increment_head(struct packet_ring_buffer *buff)
1196 {
1197 	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1198 }
1199 
1200 static void packet_inc_pending(struct packet_ring_buffer *rb)
1201 {
1202 	this_cpu_inc(*rb->pending_refcnt);
1203 }
1204 
1205 static void packet_dec_pending(struct packet_ring_buffer *rb)
1206 {
1207 	this_cpu_dec(*rb->pending_refcnt);
1208 }
1209 
1210 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1211 {
1212 	unsigned int refcnt = 0;
1213 	int cpu;
1214 
1215 	/* We don't use pending refcount in rx_ring. */
1216 	if (rb->pending_refcnt == NULL)
1217 		return 0;
1218 
1219 	for_each_possible_cpu(cpu)
1220 		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1221 
1222 	return refcnt;
1223 }
1224 
1225 static int packet_alloc_pending(struct packet_sock *po)
1226 {
1227 	po->rx_ring.pending_refcnt = NULL;
1228 
1229 	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1230 	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1231 		return -ENOBUFS;
1232 
1233 	return 0;
1234 }
1235 
1236 static void packet_free_pending(struct packet_sock *po)
1237 {
1238 	free_percpu(po->tx_ring.pending_refcnt);
1239 }
1240 
1241 #define ROOM_POW_OFF	2
1242 #define ROOM_NONE	0x0
1243 #define ROOM_LOW	0x1
1244 #define ROOM_NORMAL	0x2
1245 
1246 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1247 {
1248 	int idx, len;
1249 
1250 	len = po->rx_ring.frame_max + 1;
1251 	idx = po->rx_ring.head;
1252 	if (pow_off)
1253 		idx += len >> pow_off;
1254 	if (idx >= len)
1255 		idx -= len;
1256 	return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1257 }
1258 
1259 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1260 {
1261 	int idx, len;
1262 
1263 	len = po->rx_ring.prb_bdqc.knum_blocks;
1264 	idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1265 	if (pow_off)
1266 		idx += len >> pow_off;
1267 	if (idx >= len)
1268 		idx -= len;
1269 	return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1270 }
1271 
1272 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1273 {
1274 	struct sock *sk = &po->sk;
1275 	int ret = ROOM_NONE;
1276 
1277 	if (po->prot_hook.func != tpacket_rcv) {
1278 		int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1279 					  - (skb ? skb->truesize : 0);
1280 		if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1281 			return ROOM_NORMAL;
1282 		else if (avail > 0)
1283 			return ROOM_LOW;
1284 		else
1285 			return ROOM_NONE;
1286 	}
1287 
1288 	if (po->tp_version == TPACKET_V3) {
1289 		if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1290 			ret = ROOM_NORMAL;
1291 		else if (__tpacket_v3_has_room(po, 0))
1292 			ret = ROOM_LOW;
1293 	} else {
1294 		if (__tpacket_has_room(po, ROOM_POW_OFF))
1295 			ret = ROOM_NORMAL;
1296 		else if (__tpacket_has_room(po, 0))
1297 			ret = ROOM_LOW;
1298 	}
1299 
1300 	return ret;
1301 }
1302 
1303 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1304 {
1305 	int ret;
1306 	bool has_room;
1307 
1308 	spin_lock_bh(&po->sk.sk_receive_queue.lock);
1309 	ret = __packet_rcv_has_room(po, skb);
1310 	has_room = ret == ROOM_NORMAL;
1311 	if (po->pressure == has_room)
1312 		po->pressure = !has_room;
1313 	spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1314 
1315 	return ret;
1316 }
1317 
1318 static void packet_sock_destruct(struct sock *sk)
1319 {
1320 	skb_queue_purge(&sk->sk_error_queue);
1321 
1322 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1323 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1324 
1325 	if (!sock_flag(sk, SOCK_DEAD)) {
1326 		pr_err("Attempt to release alive packet socket: %p\n", sk);
1327 		return;
1328 	}
1329 
1330 	sk_refcnt_debug_dec(sk);
1331 }
1332 
1333 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1334 {
1335 	u32 rxhash;
1336 	int i, count = 0;
1337 
1338 	rxhash = skb_get_hash(skb);
1339 	for (i = 0; i < ROLLOVER_HLEN; i++)
1340 		if (po->rollover->history[i] == rxhash)
1341 			count++;
1342 
1343 	po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1344 	return count > (ROLLOVER_HLEN >> 1);
1345 }
1346 
1347 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1348 				      struct sk_buff *skb,
1349 				      unsigned int num)
1350 {
1351 	return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1352 }
1353 
1354 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1355 				    struct sk_buff *skb,
1356 				    unsigned int num)
1357 {
1358 	unsigned int val = atomic_inc_return(&f->rr_cur);
1359 
1360 	return val % num;
1361 }
1362 
1363 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1364 				     struct sk_buff *skb,
1365 				     unsigned int num)
1366 {
1367 	return smp_processor_id() % num;
1368 }
1369 
1370 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1371 				     struct sk_buff *skb,
1372 				     unsigned int num)
1373 {
1374 	return prandom_u32_max(num);
1375 }
1376 
1377 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1378 					  struct sk_buff *skb,
1379 					  unsigned int idx, bool try_self,
1380 					  unsigned int num)
1381 {
1382 	struct packet_sock *po, *po_next, *po_skip = NULL;
1383 	unsigned int i, j, room = ROOM_NONE;
1384 
1385 	po = pkt_sk(f->arr[idx]);
1386 
1387 	if (try_self) {
1388 		room = packet_rcv_has_room(po, skb);
1389 		if (room == ROOM_NORMAL ||
1390 		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1391 			return idx;
1392 		po_skip = po;
1393 	}
1394 
1395 	i = j = min_t(int, po->rollover->sock, num - 1);
1396 	do {
1397 		po_next = pkt_sk(f->arr[i]);
1398 		if (po_next != po_skip && !po_next->pressure &&
1399 		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1400 			if (i != j)
1401 				po->rollover->sock = i;
1402 			atomic_long_inc(&po->rollover->num);
1403 			if (room == ROOM_LOW)
1404 				atomic_long_inc(&po->rollover->num_huge);
1405 			return i;
1406 		}
1407 
1408 		if (++i == num)
1409 			i = 0;
1410 	} while (i != j);
1411 
1412 	atomic_long_inc(&po->rollover->num_failed);
1413 	return idx;
1414 }
1415 
1416 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1417 				    struct sk_buff *skb,
1418 				    unsigned int num)
1419 {
1420 	return skb_get_queue_mapping(skb) % num;
1421 }
1422 
1423 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1424 				     struct sk_buff *skb,
1425 				     unsigned int num)
1426 {
1427 	struct bpf_prog *prog;
1428 	unsigned int ret = 0;
1429 
1430 	rcu_read_lock();
1431 	prog = rcu_dereference(f->bpf_prog);
1432 	if (prog)
1433 		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1434 	rcu_read_unlock();
1435 
1436 	return ret;
1437 }
1438 
1439 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1440 {
1441 	return f->flags & (flag >> 8);
1442 }
1443 
1444 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1445 			     struct packet_type *pt, struct net_device *orig_dev)
1446 {
1447 	struct packet_fanout *f = pt->af_packet_priv;
1448 	unsigned int num = READ_ONCE(f->num_members);
1449 	struct net *net = read_pnet(&f->net);
1450 	struct packet_sock *po;
1451 	unsigned int idx;
1452 
1453 	if (!net_eq(dev_net(dev), net) || !num) {
1454 		kfree_skb(skb);
1455 		return 0;
1456 	}
1457 
1458 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1459 		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1460 		if (!skb)
1461 			return 0;
1462 	}
1463 	switch (f->type) {
1464 	case PACKET_FANOUT_HASH:
1465 	default:
1466 		idx = fanout_demux_hash(f, skb, num);
1467 		break;
1468 	case PACKET_FANOUT_LB:
1469 		idx = fanout_demux_lb(f, skb, num);
1470 		break;
1471 	case PACKET_FANOUT_CPU:
1472 		idx = fanout_demux_cpu(f, skb, num);
1473 		break;
1474 	case PACKET_FANOUT_RND:
1475 		idx = fanout_demux_rnd(f, skb, num);
1476 		break;
1477 	case PACKET_FANOUT_QM:
1478 		idx = fanout_demux_qm(f, skb, num);
1479 		break;
1480 	case PACKET_FANOUT_ROLLOVER:
1481 		idx = fanout_demux_rollover(f, skb, 0, false, num);
1482 		break;
1483 	case PACKET_FANOUT_CBPF:
1484 	case PACKET_FANOUT_EBPF:
1485 		idx = fanout_demux_bpf(f, skb, num);
1486 		break;
1487 	}
1488 
1489 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1490 		idx = fanout_demux_rollover(f, skb, idx, true, num);
1491 
1492 	po = pkt_sk(f->arr[idx]);
1493 	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1494 }
1495 
1496 DEFINE_MUTEX(fanout_mutex);
1497 EXPORT_SYMBOL_GPL(fanout_mutex);
1498 static LIST_HEAD(fanout_list);
1499 
1500 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1501 {
1502 	struct packet_fanout *f = po->fanout;
1503 
1504 	spin_lock(&f->lock);
1505 	f->arr[f->num_members] = sk;
1506 	smp_wmb();
1507 	f->num_members++;
1508 	if (f->num_members == 1)
1509 		dev_add_pack(&f->prot_hook);
1510 	spin_unlock(&f->lock);
1511 }
1512 
1513 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1514 {
1515 	struct packet_fanout *f = po->fanout;
1516 	int i;
1517 
1518 	spin_lock(&f->lock);
1519 	for (i = 0; i < f->num_members; i++) {
1520 		if (f->arr[i] == sk)
1521 			break;
1522 	}
1523 	BUG_ON(i >= f->num_members);
1524 	f->arr[i] = f->arr[f->num_members - 1];
1525 	f->num_members--;
1526 	if (f->num_members == 0)
1527 		__dev_remove_pack(&f->prot_hook);
1528 	spin_unlock(&f->lock);
1529 }
1530 
1531 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1532 {
1533 	if (sk->sk_family != PF_PACKET)
1534 		return false;
1535 
1536 	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1537 }
1538 
1539 static void fanout_init_data(struct packet_fanout *f)
1540 {
1541 	switch (f->type) {
1542 	case PACKET_FANOUT_LB:
1543 		atomic_set(&f->rr_cur, 0);
1544 		break;
1545 	case PACKET_FANOUT_CBPF:
1546 	case PACKET_FANOUT_EBPF:
1547 		RCU_INIT_POINTER(f->bpf_prog, NULL);
1548 		break;
1549 	}
1550 }
1551 
1552 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1553 {
1554 	struct bpf_prog *old;
1555 
1556 	spin_lock(&f->lock);
1557 	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1558 	rcu_assign_pointer(f->bpf_prog, new);
1559 	spin_unlock(&f->lock);
1560 
1561 	if (old) {
1562 		synchronize_net();
1563 		bpf_prog_destroy(old);
1564 	}
1565 }
1566 
1567 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1568 				unsigned int len)
1569 {
1570 	struct bpf_prog *new;
1571 	struct sock_fprog fprog;
1572 	int ret;
1573 
1574 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1575 		return -EPERM;
1576 	if (len != sizeof(fprog))
1577 		return -EINVAL;
1578 	if (copy_from_user(&fprog, data, len))
1579 		return -EFAULT;
1580 
1581 	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1582 	if (ret)
1583 		return ret;
1584 
1585 	__fanout_set_data_bpf(po->fanout, new);
1586 	return 0;
1587 }
1588 
1589 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1590 				unsigned int len)
1591 {
1592 	struct bpf_prog *new;
1593 	u32 fd;
1594 
1595 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1596 		return -EPERM;
1597 	if (len != sizeof(fd))
1598 		return -EINVAL;
1599 	if (copy_from_user(&fd, data, len))
1600 		return -EFAULT;
1601 
1602 	new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1603 	if (IS_ERR(new))
1604 		return PTR_ERR(new);
1605 
1606 	__fanout_set_data_bpf(po->fanout, new);
1607 	return 0;
1608 }
1609 
1610 static int fanout_set_data(struct packet_sock *po, char __user *data,
1611 			   unsigned int len)
1612 {
1613 	switch (po->fanout->type) {
1614 	case PACKET_FANOUT_CBPF:
1615 		return fanout_set_data_cbpf(po, data, len);
1616 	case PACKET_FANOUT_EBPF:
1617 		return fanout_set_data_ebpf(po, data, len);
1618 	default:
1619 		return -EINVAL;
1620 	};
1621 }
1622 
1623 static void fanout_release_data(struct packet_fanout *f)
1624 {
1625 	switch (f->type) {
1626 	case PACKET_FANOUT_CBPF:
1627 	case PACKET_FANOUT_EBPF:
1628 		__fanout_set_data_bpf(f, NULL);
1629 	};
1630 }
1631 
1632 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1633 {
1634 	struct packet_rollover *rollover = NULL;
1635 	struct packet_sock *po = pkt_sk(sk);
1636 	struct packet_fanout *f, *match;
1637 	u8 type = type_flags & 0xff;
1638 	u8 flags = type_flags >> 8;
1639 	int err;
1640 
1641 	switch (type) {
1642 	case PACKET_FANOUT_ROLLOVER:
1643 		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1644 			return -EINVAL;
1645 	case PACKET_FANOUT_HASH:
1646 	case PACKET_FANOUT_LB:
1647 	case PACKET_FANOUT_CPU:
1648 	case PACKET_FANOUT_RND:
1649 	case PACKET_FANOUT_QM:
1650 	case PACKET_FANOUT_CBPF:
1651 	case PACKET_FANOUT_EBPF:
1652 		break;
1653 	default:
1654 		return -EINVAL;
1655 	}
1656 
1657 	mutex_lock(&fanout_mutex);
1658 
1659 	err = -EINVAL;
1660 	if (!po->running)
1661 		goto out;
1662 
1663 	err = -EALREADY;
1664 	if (po->fanout)
1665 		goto out;
1666 
1667 	if (type == PACKET_FANOUT_ROLLOVER ||
1668 	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1669 		err = -ENOMEM;
1670 		rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1671 		if (!rollover)
1672 			goto out;
1673 		atomic_long_set(&rollover->num, 0);
1674 		atomic_long_set(&rollover->num_huge, 0);
1675 		atomic_long_set(&rollover->num_failed, 0);
1676 		po->rollover = rollover;
1677 	}
1678 
1679 	match = NULL;
1680 	list_for_each_entry(f, &fanout_list, list) {
1681 		if (f->id == id &&
1682 		    read_pnet(&f->net) == sock_net(sk)) {
1683 			match = f;
1684 			break;
1685 		}
1686 	}
1687 	err = -EINVAL;
1688 	if (match && match->flags != flags)
1689 		goto out;
1690 	if (!match) {
1691 		err = -ENOMEM;
1692 		match = kzalloc(sizeof(*match), GFP_KERNEL);
1693 		if (!match)
1694 			goto out;
1695 		write_pnet(&match->net, sock_net(sk));
1696 		match->id = id;
1697 		match->type = type;
1698 		match->flags = flags;
1699 		INIT_LIST_HEAD(&match->list);
1700 		spin_lock_init(&match->lock);
1701 		atomic_set(&match->sk_ref, 0);
1702 		fanout_init_data(match);
1703 		match->prot_hook.type = po->prot_hook.type;
1704 		match->prot_hook.dev = po->prot_hook.dev;
1705 		match->prot_hook.func = packet_rcv_fanout;
1706 		match->prot_hook.af_packet_priv = match;
1707 		match->prot_hook.id_match = match_fanout_group;
1708 		list_add(&match->list, &fanout_list);
1709 	}
1710 	err = -EINVAL;
1711 	if (match->type == type &&
1712 	    match->prot_hook.type == po->prot_hook.type &&
1713 	    match->prot_hook.dev == po->prot_hook.dev) {
1714 		err = -ENOSPC;
1715 		if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1716 			__dev_remove_pack(&po->prot_hook);
1717 			po->fanout = match;
1718 			atomic_inc(&match->sk_ref);
1719 			__fanout_link(sk, po);
1720 			err = 0;
1721 		}
1722 	}
1723 out:
1724 	if (err && rollover) {
1725 		kfree(rollover);
1726 		po->rollover = NULL;
1727 	}
1728 	mutex_unlock(&fanout_mutex);
1729 	return err;
1730 }
1731 
1732 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1733  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1734  * It is the responsibility of the caller to call fanout_release_data() and
1735  * free the returned packet_fanout (after synchronize_net())
1736  */
1737 static struct packet_fanout *fanout_release(struct sock *sk)
1738 {
1739 	struct packet_sock *po = pkt_sk(sk);
1740 	struct packet_fanout *f;
1741 
1742 	mutex_lock(&fanout_mutex);
1743 	f = po->fanout;
1744 	if (f) {
1745 		po->fanout = NULL;
1746 
1747 		if (atomic_dec_and_test(&f->sk_ref))
1748 			list_del(&f->list);
1749 		else
1750 			f = NULL;
1751 
1752 		if (po->rollover)
1753 			kfree_rcu(po->rollover, rcu);
1754 	}
1755 	mutex_unlock(&fanout_mutex);
1756 
1757 	return f;
1758 }
1759 
1760 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1761 					  struct sk_buff *skb)
1762 {
1763 	/* Earlier code assumed this would be a VLAN pkt, double-check
1764 	 * this now that we have the actual packet in hand. We can only
1765 	 * do this check on Ethernet devices.
1766 	 */
1767 	if (unlikely(dev->type != ARPHRD_ETHER))
1768 		return false;
1769 
1770 	skb_reset_mac_header(skb);
1771 	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1772 }
1773 
1774 static const struct proto_ops packet_ops;
1775 
1776 static const struct proto_ops packet_ops_spkt;
1777 
1778 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1779 			   struct packet_type *pt, struct net_device *orig_dev)
1780 {
1781 	struct sock *sk;
1782 	struct sockaddr_pkt *spkt;
1783 
1784 	/*
1785 	 *	When we registered the protocol we saved the socket in the data
1786 	 *	field for just this event.
1787 	 */
1788 
1789 	sk = pt->af_packet_priv;
1790 
1791 	/*
1792 	 *	Yank back the headers [hope the device set this
1793 	 *	right or kerboom...]
1794 	 *
1795 	 *	Incoming packets have ll header pulled,
1796 	 *	push it back.
1797 	 *
1798 	 *	For outgoing ones skb->data == skb_mac_header(skb)
1799 	 *	so that this procedure is noop.
1800 	 */
1801 
1802 	if (skb->pkt_type == PACKET_LOOPBACK)
1803 		goto out;
1804 
1805 	if (!net_eq(dev_net(dev), sock_net(sk)))
1806 		goto out;
1807 
1808 	skb = skb_share_check(skb, GFP_ATOMIC);
1809 	if (skb == NULL)
1810 		goto oom;
1811 
1812 	/* drop any routing info */
1813 	skb_dst_drop(skb);
1814 
1815 	/* drop conntrack reference */
1816 	nf_reset(skb);
1817 
1818 	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1819 
1820 	skb_push(skb, skb->data - skb_mac_header(skb));
1821 
1822 	/*
1823 	 *	The SOCK_PACKET socket receives _all_ frames.
1824 	 */
1825 
1826 	spkt->spkt_family = dev->type;
1827 	strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1828 	spkt->spkt_protocol = skb->protocol;
1829 
1830 	/*
1831 	 *	Charge the memory to the socket. This is done specifically
1832 	 *	to prevent sockets using all the memory up.
1833 	 */
1834 
1835 	if (sock_queue_rcv_skb(sk, skb) == 0)
1836 		return 0;
1837 
1838 out:
1839 	kfree_skb(skb);
1840 oom:
1841 	return 0;
1842 }
1843 
1844 
1845 /*
1846  *	Output a raw packet to a device layer. This bypasses all the other
1847  *	protocol layers and you must therefore supply it with a complete frame
1848  */
1849 
1850 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1851 			       size_t len)
1852 {
1853 	struct sock *sk = sock->sk;
1854 	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1855 	struct sk_buff *skb = NULL;
1856 	struct net_device *dev;
1857 	struct sockcm_cookie sockc;
1858 	__be16 proto = 0;
1859 	int err;
1860 	int extra_len = 0;
1861 
1862 	/*
1863 	 *	Get and verify the address.
1864 	 */
1865 
1866 	if (saddr) {
1867 		if (msg->msg_namelen < sizeof(struct sockaddr))
1868 			return -EINVAL;
1869 		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1870 			proto = saddr->spkt_protocol;
1871 	} else
1872 		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1873 
1874 	/*
1875 	 *	Find the device first to size check it
1876 	 */
1877 
1878 	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1879 retry:
1880 	rcu_read_lock();
1881 	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1882 	err = -ENODEV;
1883 	if (dev == NULL)
1884 		goto out_unlock;
1885 
1886 	err = -ENETDOWN;
1887 	if (!(dev->flags & IFF_UP))
1888 		goto out_unlock;
1889 
1890 	/*
1891 	 * You may not queue a frame bigger than the mtu. This is the lowest level
1892 	 * raw protocol and you must do your own fragmentation at this level.
1893 	 */
1894 
1895 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1896 		if (!netif_supports_nofcs(dev)) {
1897 			err = -EPROTONOSUPPORT;
1898 			goto out_unlock;
1899 		}
1900 		extra_len = 4; /* We're doing our own CRC */
1901 	}
1902 
1903 	err = -EMSGSIZE;
1904 	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1905 		goto out_unlock;
1906 
1907 	if (!skb) {
1908 		size_t reserved = LL_RESERVED_SPACE(dev);
1909 		int tlen = dev->needed_tailroom;
1910 		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1911 
1912 		rcu_read_unlock();
1913 		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1914 		if (skb == NULL)
1915 			return -ENOBUFS;
1916 		/* FIXME: Save some space for broken drivers that write a hard
1917 		 * header at transmission time by themselves. PPP is the notable
1918 		 * one here. This should really be fixed at the driver level.
1919 		 */
1920 		skb_reserve(skb, reserved);
1921 		skb_reset_network_header(skb);
1922 
1923 		/* Try to align data part correctly */
1924 		if (hhlen) {
1925 			skb->data -= hhlen;
1926 			skb->tail -= hhlen;
1927 			if (len < hhlen)
1928 				skb_reset_network_header(skb);
1929 		}
1930 		err = memcpy_from_msg(skb_put(skb, len), msg, len);
1931 		if (err)
1932 			goto out_free;
1933 		goto retry;
1934 	}
1935 
1936 	if (!dev_validate_header(dev, skb->data, len)) {
1937 		err = -EINVAL;
1938 		goto out_unlock;
1939 	}
1940 	if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1941 	    !packet_extra_vlan_len_allowed(dev, skb)) {
1942 		err = -EMSGSIZE;
1943 		goto out_unlock;
1944 	}
1945 
1946 	sockc.tsflags = sk->sk_tsflags;
1947 	if (msg->msg_controllen) {
1948 		err = sock_cmsg_send(sk, msg, &sockc);
1949 		if (unlikely(err))
1950 			goto out_unlock;
1951 	}
1952 
1953 	skb->protocol = proto;
1954 	skb->dev = dev;
1955 	skb->priority = sk->sk_priority;
1956 	skb->mark = sk->sk_mark;
1957 
1958 	sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1959 
1960 	if (unlikely(extra_len == 4))
1961 		skb->no_fcs = 1;
1962 
1963 	skb_probe_transport_header(skb, 0);
1964 
1965 	dev_queue_xmit(skb);
1966 	rcu_read_unlock();
1967 	return len;
1968 
1969 out_unlock:
1970 	rcu_read_unlock();
1971 out_free:
1972 	kfree_skb(skb);
1973 	return err;
1974 }
1975 
1976 static unsigned int run_filter(struct sk_buff *skb,
1977 			       const struct sock *sk,
1978 			       unsigned int res)
1979 {
1980 	struct sk_filter *filter;
1981 
1982 	rcu_read_lock();
1983 	filter = rcu_dereference(sk->sk_filter);
1984 	if (filter != NULL)
1985 		res = bpf_prog_run_clear_cb(filter->prog, skb);
1986 	rcu_read_unlock();
1987 
1988 	return res;
1989 }
1990 
1991 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
1992 			   size_t *len)
1993 {
1994 	struct virtio_net_hdr vnet_hdr;
1995 
1996 	if (*len < sizeof(vnet_hdr))
1997 		return -EINVAL;
1998 	*len -= sizeof(vnet_hdr);
1999 
2000 	if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true))
2001 		return -EINVAL;
2002 
2003 	return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2004 }
2005 
2006 /*
2007  * This function makes lazy skb cloning in hope that most of packets
2008  * are discarded by BPF.
2009  *
2010  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2011  * and skb->cb are mangled. It works because (and until) packets
2012  * falling here are owned by current CPU. Output packets are cloned
2013  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2014  * sequencially, so that if we return skb to original state on exit,
2015  * we will not harm anyone.
2016  */
2017 
2018 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2019 		      struct packet_type *pt, struct net_device *orig_dev)
2020 {
2021 	struct sock *sk;
2022 	struct sockaddr_ll *sll;
2023 	struct packet_sock *po;
2024 	u8 *skb_head = skb->data;
2025 	int skb_len = skb->len;
2026 	unsigned int snaplen, res;
2027 	bool is_drop_n_account = false;
2028 
2029 	if (skb->pkt_type == PACKET_LOOPBACK)
2030 		goto drop;
2031 
2032 	sk = pt->af_packet_priv;
2033 	po = pkt_sk(sk);
2034 
2035 	if (!net_eq(dev_net(dev), sock_net(sk)))
2036 		goto drop;
2037 
2038 	skb->dev = dev;
2039 
2040 	if (dev->header_ops) {
2041 		/* The device has an explicit notion of ll header,
2042 		 * exported to higher levels.
2043 		 *
2044 		 * Otherwise, the device hides details of its frame
2045 		 * structure, so that corresponding packet head is
2046 		 * never delivered to user.
2047 		 */
2048 		if (sk->sk_type != SOCK_DGRAM)
2049 			skb_push(skb, skb->data - skb_mac_header(skb));
2050 		else if (skb->pkt_type == PACKET_OUTGOING) {
2051 			/* Special case: outgoing packets have ll header at head */
2052 			skb_pull(skb, skb_network_offset(skb));
2053 		}
2054 	}
2055 
2056 	snaplen = skb->len;
2057 
2058 	res = run_filter(skb, sk, snaplen);
2059 	if (!res)
2060 		goto drop_n_restore;
2061 	if (snaplen > res)
2062 		snaplen = res;
2063 
2064 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2065 		goto drop_n_acct;
2066 
2067 	if (skb_shared(skb)) {
2068 		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2069 		if (nskb == NULL)
2070 			goto drop_n_acct;
2071 
2072 		if (skb_head != skb->data) {
2073 			skb->data = skb_head;
2074 			skb->len = skb_len;
2075 		}
2076 		consume_skb(skb);
2077 		skb = nskb;
2078 	}
2079 
2080 	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2081 
2082 	sll = &PACKET_SKB_CB(skb)->sa.ll;
2083 	sll->sll_hatype = dev->type;
2084 	sll->sll_pkttype = skb->pkt_type;
2085 	if (unlikely(po->origdev))
2086 		sll->sll_ifindex = orig_dev->ifindex;
2087 	else
2088 		sll->sll_ifindex = dev->ifindex;
2089 
2090 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2091 
2092 	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2093 	 * Use their space for storing the original skb length.
2094 	 */
2095 	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2096 
2097 	if (pskb_trim(skb, snaplen))
2098 		goto drop_n_acct;
2099 
2100 	skb_set_owner_r(skb, sk);
2101 	skb->dev = NULL;
2102 	skb_dst_drop(skb);
2103 
2104 	/* drop conntrack reference */
2105 	nf_reset(skb);
2106 
2107 	spin_lock(&sk->sk_receive_queue.lock);
2108 	po->stats.stats1.tp_packets++;
2109 	sock_skb_set_dropcount(sk, skb);
2110 	__skb_queue_tail(&sk->sk_receive_queue, skb);
2111 	spin_unlock(&sk->sk_receive_queue.lock);
2112 	sk->sk_data_ready(sk);
2113 	return 0;
2114 
2115 drop_n_acct:
2116 	is_drop_n_account = true;
2117 	spin_lock(&sk->sk_receive_queue.lock);
2118 	po->stats.stats1.tp_drops++;
2119 	atomic_inc(&sk->sk_drops);
2120 	spin_unlock(&sk->sk_receive_queue.lock);
2121 
2122 drop_n_restore:
2123 	if (skb_head != skb->data && skb_shared(skb)) {
2124 		skb->data = skb_head;
2125 		skb->len = skb_len;
2126 	}
2127 drop:
2128 	if (!is_drop_n_account)
2129 		consume_skb(skb);
2130 	else
2131 		kfree_skb(skb);
2132 	return 0;
2133 }
2134 
2135 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2136 		       struct packet_type *pt, struct net_device *orig_dev)
2137 {
2138 	struct sock *sk;
2139 	struct packet_sock *po;
2140 	struct sockaddr_ll *sll;
2141 	union tpacket_uhdr h;
2142 	u8 *skb_head = skb->data;
2143 	int skb_len = skb->len;
2144 	unsigned int snaplen, res;
2145 	unsigned long status = TP_STATUS_USER;
2146 	unsigned short macoff, netoff, hdrlen;
2147 	struct sk_buff *copy_skb = NULL;
2148 	struct timespec ts;
2149 	__u32 ts_status;
2150 	bool is_drop_n_account = false;
2151 
2152 	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2153 	 * We may add members to them until current aligned size without forcing
2154 	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2155 	 */
2156 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2157 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2158 
2159 	if (skb->pkt_type == PACKET_LOOPBACK)
2160 		goto drop;
2161 
2162 	sk = pt->af_packet_priv;
2163 	po = pkt_sk(sk);
2164 
2165 	if (!net_eq(dev_net(dev), sock_net(sk)))
2166 		goto drop;
2167 
2168 	if (dev->header_ops) {
2169 		if (sk->sk_type != SOCK_DGRAM)
2170 			skb_push(skb, skb->data - skb_mac_header(skb));
2171 		else if (skb->pkt_type == PACKET_OUTGOING) {
2172 			/* Special case: outgoing packets have ll header at head */
2173 			skb_pull(skb, skb_network_offset(skb));
2174 		}
2175 	}
2176 
2177 	snaplen = skb->len;
2178 
2179 	res = run_filter(skb, sk, snaplen);
2180 	if (!res)
2181 		goto drop_n_restore;
2182 
2183 	if (skb->ip_summed == CHECKSUM_PARTIAL)
2184 		status |= TP_STATUS_CSUMNOTREADY;
2185 	else if (skb->pkt_type != PACKET_OUTGOING &&
2186 		 (skb->ip_summed == CHECKSUM_COMPLETE ||
2187 		  skb_csum_unnecessary(skb)))
2188 		status |= TP_STATUS_CSUM_VALID;
2189 
2190 	if (snaplen > res)
2191 		snaplen = res;
2192 
2193 	if (sk->sk_type == SOCK_DGRAM) {
2194 		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2195 				  po->tp_reserve;
2196 	} else {
2197 		unsigned int maclen = skb_network_offset(skb);
2198 		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2199 				       (maclen < 16 ? 16 : maclen)) +
2200 				       po->tp_reserve;
2201 		if (po->has_vnet_hdr)
2202 			netoff += sizeof(struct virtio_net_hdr);
2203 		macoff = netoff - maclen;
2204 	}
2205 	if (po->tp_version <= TPACKET_V2) {
2206 		if (macoff + snaplen > po->rx_ring.frame_size) {
2207 			if (po->copy_thresh &&
2208 			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2209 				if (skb_shared(skb)) {
2210 					copy_skb = skb_clone(skb, GFP_ATOMIC);
2211 				} else {
2212 					copy_skb = skb_get(skb);
2213 					skb_head = skb->data;
2214 				}
2215 				if (copy_skb)
2216 					skb_set_owner_r(copy_skb, sk);
2217 			}
2218 			snaplen = po->rx_ring.frame_size - macoff;
2219 			if ((int)snaplen < 0)
2220 				snaplen = 0;
2221 		}
2222 	} else if (unlikely(macoff + snaplen >
2223 			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2224 		u32 nval;
2225 
2226 		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2227 		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2228 			    snaplen, nval, macoff);
2229 		snaplen = nval;
2230 		if (unlikely((int)snaplen < 0)) {
2231 			snaplen = 0;
2232 			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2233 		}
2234 	}
2235 	spin_lock(&sk->sk_receive_queue.lock);
2236 	h.raw = packet_current_rx_frame(po, skb,
2237 					TP_STATUS_KERNEL, (macoff+snaplen));
2238 	if (!h.raw)
2239 		goto drop_n_account;
2240 	if (po->tp_version <= TPACKET_V2) {
2241 		packet_increment_rx_head(po, &po->rx_ring);
2242 	/*
2243 	 * LOSING will be reported till you read the stats,
2244 	 * because it's COR - Clear On Read.
2245 	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2246 	 * at packet level.
2247 	 */
2248 		if (po->stats.stats1.tp_drops)
2249 			status |= TP_STATUS_LOSING;
2250 	}
2251 	po->stats.stats1.tp_packets++;
2252 	if (copy_skb) {
2253 		status |= TP_STATUS_COPY;
2254 		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2255 	}
2256 	spin_unlock(&sk->sk_receive_queue.lock);
2257 
2258 	if (po->has_vnet_hdr) {
2259 		if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2260 					    sizeof(struct virtio_net_hdr),
2261 					    vio_le(), true)) {
2262 			spin_lock(&sk->sk_receive_queue.lock);
2263 			goto drop_n_account;
2264 		}
2265 	}
2266 
2267 	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2268 
2269 	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2270 		getnstimeofday(&ts);
2271 
2272 	status |= ts_status;
2273 
2274 	switch (po->tp_version) {
2275 	case TPACKET_V1:
2276 		h.h1->tp_len = skb->len;
2277 		h.h1->tp_snaplen = snaplen;
2278 		h.h1->tp_mac = macoff;
2279 		h.h1->tp_net = netoff;
2280 		h.h1->tp_sec = ts.tv_sec;
2281 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2282 		hdrlen = sizeof(*h.h1);
2283 		break;
2284 	case TPACKET_V2:
2285 		h.h2->tp_len = skb->len;
2286 		h.h2->tp_snaplen = snaplen;
2287 		h.h2->tp_mac = macoff;
2288 		h.h2->tp_net = netoff;
2289 		h.h2->tp_sec = ts.tv_sec;
2290 		h.h2->tp_nsec = ts.tv_nsec;
2291 		if (skb_vlan_tag_present(skb)) {
2292 			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2293 			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2294 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2295 		} else {
2296 			h.h2->tp_vlan_tci = 0;
2297 			h.h2->tp_vlan_tpid = 0;
2298 		}
2299 		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2300 		hdrlen = sizeof(*h.h2);
2301 		break;
2302 	case TPACKET_V3:
2303 		/* tp_nxt_offset,vlan are already populated above.
2304 		 * So DONT clear those fields here
2305 		 */
2306 		h.h3->tp_status |= status;
2307 		h.h3->tp_len = skb->len;
2308 		h.h3->tp_snaplen = snaplen;
2309 		h.h3->tp_mac = macoff;
2310 		h.h3->tp_net = netoff;
2311 		h.h3->tp_sec  = ts.tv_sec;
2312 		h.h3->tp_nsec = ts.tv_nsec;
2313 		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2314 		hdrlen = sizeof(*h.h3);
2315 		break;
2316 	default:
2317 		BUG();
2318 	}
2319 
2320 	sll = h.raw + TPACKET_ALIGN(hdrlen);
2321 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2322 	sll->sll_family = AF_PACKET;
2323 	sll->sll_hatype = dev->type;
2324 	sll->sll_protocol = skb->protocol;
2325 	sll->sll_pkttype = skb->pkt_type;
2326 	if (unlikely(po->origdev))
2327 		sll->sll_ifindex = orig_dev->ifindex;
2328 	else
2329 		sll->sll_ifindex = dev->ifindex;
2330 
2331 	smp_mb();
2332 
2333 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2334 	if (po->tp_version <= TPACKET_V2) {
2335 		u8 *start, *end;
2336 
2337 		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2338 					macoff + snaplen);
2339 
2340 		for (start = h.raw; start < end; start += PAGE_SIZE)
2341 			flush_dcache_page(pgv_to_page(start));
2342 	}
2343 	smp_wmb();
2344 #endif
2345 
2346 	if (po->tp_version <= TPACKET_V2) {
2347 		__packet_set_status(po, h.raw, status);
2348 		sk->sk_data_ready(sk);
2349 	} else {
2350 		prb_clear_blk_fill_status(&po->rx_ring);
2351 	}
2352 
2353 drop_n_restore:
2354 	if (skb_head != skb->data && skb_shared(skb)) {
2355 		skb->data = skb_head;
2356 		skb->len = skb_len;
2357 	}
2358 drop:
2359 	if (!is_drop_n_account)
2360 		consume_skb(skb);
2361 	else
2362 		kfree_skb(skb);
2363 	return 0;
2364 
2365 drop_n_account:
2366 	is_drop_n_account = true;
2367 	po->stats.stats1.tp_drops++;
2368 	spin_unlock(&sk->sk_receive_queue.lock);
2369 
2370 	sk->sk_data_ready(sk);
2371 	kfree_skb(copy_skb);
2372 	goto drop_n_restore;
2373 }
2374 
2375 static void tpacket_destruct_skb(struct sk_buff *skb)
2376 {
2377 	struct packet_sock *po = pkt_sk(skb->sk);
2378 
2379 	if (likely(po->tx_ring.pg_vec)) {
2380 		void *ph;
2381 		__u32 ts;
2382 
2383 		ph = skb_shinfo(skb)->destructor_arg;
2384 		packet_dec_pending(&po->tx_ring);
2385 
2386 		ts = __packet_set_timestamp(po, ph, skb);
2387 		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2388 	}
2389 
2390 	sock_wfree(skb);
2391 }
2392 
2393 static void tpacket_set_protocol(const struct net_device *dev,
2394 				 struct sk_buff *skb)
2395 {
2396 	if (dev->type == ARPHRD_ETHER) {
2397 		skb_reset_mac_header(skb);
2398 		skb->protocol = eth_hdr(skb)->h_proto;
2399 	}
2400 }
2401 
2402 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2403 {
2404 	if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2405 	    (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2406 	     __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2407 	      __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2408 		vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2409 			 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2410 			__virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2411 
2412 	if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2413 		return -EINVAL;
2414 
2415 	return 0;
2416 }
2417 
2418 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2419 				 struct virtio_net_hdr *vnet_hdr)
2420 {
2421 	if (*len < sizeof(*vnet_hdr))
2422 		return -EINVAL;
2423 	*len -= sizeof(*vnet_hdr);
2424 
2425 	if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2426 		return -EFAULT;
2427 
2428 	return __packet_snd_vnet_parse(vnet_hdr, *len);
2429 }
2430 
2431 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2432 		void *frame, struct net_device *dev, void *data, int tp_len,
2433 		__be16 proto, unsigned char *addr, int hlen, int copylen,
2434 		const struct sockcm_cookie *sockc)
2435 {
2436 	union tpacket_uhdr ph;
2437 	int to_write, offset, len, nr_frags, len_max;
2438 	struct socket *sock = po->sk.sk_socket;
2439 	struct page *page;
2440 	int err;
2441 
2442 	ph.raw = frame;
2443 
2444 	skb->protocol = proto;
2445 	skb->dev = dev;
2446 	skb->priority = po->sk.sk_priority;
2447 	skb->mark = po->sk.sk_mark;
2448 	sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2449 	skb_shinfo(skb)->destructor_arg = ph.raw;
2450 
2451 	skb_reserve(skb, hlen);
2452 	skb_reset_network_header(skb);
2453 
2454 	to_write = tp_len;
2455 
2456 	if (sock->type == SOCK_DGRAM) {
2457 		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2458 				NULL, tp_len);
2459 		if (unlikely(err < 0))
2460 			return -EINVAL;
2461 	} else if (copylen) {
2462 		int hdrlen = min_t(int, copylen, tp_len);
2463 
2464 		skb_push(skb, dev->hard_header_len);
2465 		skb_put(skb, copylen - dev->hard_header_len);
2466 		err = skb_store_bits(skb, 0, data, hdrlen);
2467 		if (unlikely(err))
2468 			return err;
2469 		if (!dev_validate_header(dev, skb->data, hdrlen))
2470 			return -EINVAL;
2471 		if (!skb->protocol)
2472 			tpacket_set_protocol(dev, skb);
2473 
2474 		data += hdrlen;
2475 		to_write -= hdrlen;
2476 	}
2477 
2478 	offset = offset_in_page(data);
2479 	len_max = PAGE_SIZE - offset;
2480 	len = ((to_write > len_max) ? len_max : to_write);
2481 
2482 	skb->data_len = to_write;
2483 	skb->len += to_write;
2484 	skb->truesize += to_write;
2485 	atomic_add(to_write, &po->sk.sk_wmem_alloc);
2486 
2487 	while (likely(to_write)) {
2488 		nr_frags = skb_shinfo(skb)->nr_frags;
2489 
2490 		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2491 			pr_err("Packet exceed the number of skb frags(%lu)\n",
2492 			       MAX_SKB_FRAGS);
2493 			return -EFAULT;
2494 		}
2495 
2496 		page = pgv_to_page(data);
2497 		data += len;
2498 		flush_dcache_page(page);
2499 		get_page(page);
2500 		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2501 		to_write -= len;
2502 		offset = 0;
2503 		len_max = PAGE_SIZE;
2504 		len = ((to_write > len_max) ? len_max : to_write);
2505 	}
2506 
2507 	skb_probe_transport_header(skb, 0);
2508 
2509 	return tp_len;
2510 }
2511 
2512 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2513 				int size_max, void **data)
2514 {
2515 	union tpacket_uhdr ph;
2516 	int tp_len, off;
2517 
2518 	ph.raw = frame;
2519 
2520 	switch (po->tp_version) {
2521 	case TPACKET_V3:
2522 		if (ph.h3->tp_next_offset != 0) {
2523 			pr_warn_once("variable sized slot not supported");
2524 			return -EINVAL;
2525 		}
2526 		tp_len = ph.h3->tp_len;
2527 		break;
2528 	case TPACKET_V2:
2529 		tp_len = ph.h2->tp_len;
2530 		break;
2531 	default:
2532 		tp_len = ph.h1->tp_len;
2533 		break;
2534 	}
2535 	if (unlikely(tp_len > size_max)) {
2536 		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2537 		return -EMSGSIZE;
2538 	}
2539 
2540 	if (unlikely(po->tp_tx_has_off)) {
2541 		int off_min, off_max;
2542 
2543 		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2544 		off_max = po->tx_ring.frame_size - tp_len;
2545 		if (po->sk.sk_type == SOCK_DGRAM) {
2546 			switch (po->tp_version) {
2547 			case TPACKET_V3:
2548 				off = ph.h3->tp_net;
2549 				break;
2550 			case TPACKET_V2:
2551 				off = ph.h2->tp_net;
2552 				break;
2553 			default:
2554 				off = ph.h1->tp_net;
2555 				break;
2556 			}
2557 		} else {
2558 			switch (po->tp_version) {
2559 			case TPACKET_V3:
2560 				off = ph.h3->tp_mac;
2561 				break;
2562 			case TPACKET_V2:
2563 				off = ph.h2->tp_mac;
2564 				break;
2565 			default:
2566 				off = ph.h1->tp_mac;
2567 				break;
2568 			}
2569 		}
2570 		if (unlikely((off < off_min) || (off_max < off)))
2571 			return -EINVAL;
2572 	} else {
2573 		off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2574 	}
2575 
2576 	*data = frame + off;
2577 	return tp_len;
2578 }
2579 
2580 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2581 {
2582 	struct sk_buff *skb;
2583 	struct net_device *dev;
2584 	struct virtio_net_hdr *vnet_hdr = NULL;
2585 	struct sockcm_cookie sockc;
2586 	__be16 proto;
2587 	int err, reserve = 0;
2588 	void *ph;
2589 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2590 	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2591 	int tp_len, size_max;
2592 	unsigned char *addr;
2593 	void *data;
2594 	int len_sum = 0;
2595 	int status = TP_STATUS_AVAILABLE;
2596 	int hlen, tlen, copylen = 0;
2597 
2598 	mutex_lock(&po->pg_vec_lock);
2599 
2600 	if (likely(saddr == NULL)) {
2601 		dev	= packet_cached_dev_get(po);
2602 		proto	= po->num;
2603 		addr	= NULL;
2604 	} else {
2605 		err = -EINVAL;
2606 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2607 			goto out;
2608 		if (msg->msg_namelen < (saddr->sll_halen
2609 					+ offsetof(struct sockaddr_ll,
2610 						sll_addr)))
2611 			goto out;
2612 		proto	= saddr->sll_protocol;
2613 		addr	= saddr->sll_addr;
2614 		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2615 	}
2616 
2617 	sockc.tsflags = po->sk.sk_tsflags;
2618 	if (msg->msg_controllen) {
2619 		err = sock_cmsg_send(&po->sk, msg, &sockc);
2620 		if (unlikely(err))
2621 			goto out;
2622 	}
2623 
2624 	err = -ENXIO;
2625 	if (unlikely(dev == NULL))
2626 		goto out;
2627 	err = -ENETDOWN;
2628 	if (unlikely(!(dev->flags & IFF_UP)))
2629 		goto out_put;
2630 
2631 	if (po->sk.sk_socket->type == SOCK_RAW)
2632 		reserve = dev->hard_header_len;
2633 	size_max = po->tx_ring.frame_size
2634 		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2635 
2636 	if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2637 		size_max = dev->mtu + reserve + VLAN_HLEN;
2638 
2639 	do {
2640 		ph = packet_current_frame(po, &po->tx_ring,
2641 					  TP_STATUS_SEND_REQUEST);
2642 		if (unlikely(ph == NULL)) {
2643 			if (need_wait && need_resched())
2644 				schedule();
2645 			continue;
2646 		}
2647 
2648 		skb = NULL;
2649 		tp_len = tpacket_parse_header(po, ph, size_max, &data);
2650 		if (tp_len < 0)
2651 			goto tpacket_error;
2652 
2653 		status = TP_STATUS_SEND_REQUEST;
2654 		hlen = LL_RESERVED_SPACE(dev);
2655 		tlen = dev->needed_tailroom;
2656 		if (po->has_vnet_hdr) {
2657 			vnet_hdr = data;
2658 			data += sizeof(*vnet_hdr);
2659 			tp_len -= sizeof(*vnet_hdr);
2660 			if (tp_len < 0 ||
2661 			    __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2662 				tp_len = -EINVAL;
2663 				goto tpacket_error;
2664 			}
2665 			copylen = __virtio16_to_cpu(vio_le(),
2666 						    vnet_hdr->hdr_len);
2667 		}
2668 		copylen = max_t(int, copylen, dev->hard_header_len);
2669 		skb = sock_alloc_send_skb(&po->sk,
2670 				hlen + tlen + sizeof(struct sockaddr_ll) +
2671 				(copylen - dev->hard_header_len),
2672 				!need_wait, &err);
2673 
2674 		if (unlikely(skb == NULL)) {
2675 			/* we assume the socket was initially writeable ... */
2676 			if (likely(len_sum > 0))
2677 				err = len_sum;
2678 			goto out_status;
2679 		}
2680 		tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2681 					  addr, hlen, copylen, &sockc);
2682 		if (likely(tp_len >= 0) &&
2683 		    tp_len > dev->mtu + reserve &&
2684 		    !po->has_vnet_hdr &&
2685 		    !packet_extra_vlan_len_allowed(dev, skb))
2686 			tp_len = -EMSGSIZE;
2687 
2688 		if (unlikely(tp_len < 0)) {
2689 tpacket_error:
2690 			if (po->tp_loss) {
2691 				__packet_set_status(po, ph,
2692 						TP_STATUS_AVAILABLE);
2693 				packet_increment_head(&po->tx_ring);
2694 				kfree_skb(skb);
2695 				continue;
2696 			} else {
2697 				status = TP_STATUS_WRONG_FORMAT;
2698 				err = tp_len;
2699 				goto out_status;
2700 			}
2701 		}
2702 
2703 		if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2704 							      vio_le())) {
2705 			tp_len = -EINVAL;
2706 			goto tpacket_error;
2707 		}
2708 
2709 		packet_pick_tx_queue(dev, skb);
2710 
2711 		skb->destructor = tpacket_destruct_skb;
2712 		__packet_set_status(po, ph, TP_STATUS_SENDING);
2713 		packet_inc_pending(&po->tx_ring);
2714 
2715 		status = TP_STATUS_SEND_REQUEST;
2716 		err = po->xmit(skb);
2717 		if (unlikely(err > 0)) {
2718 			err = net_xmit_errno(err);
2719 			if (err && __packet_get_status(po, ph) ==
2720 				   TP_STATUS_AVAILABLE) {
2721 				/* skb was destructed already */
2722 				skb = NULL;
2723 				goto out_status;
2724 			}
2725 			/*
2726 			 * skb was dropped but not destructed yet;
2727 			 * let's treat it like congestion or err < 0
2728 			 */
2729 			err = 0;
2730 		}
2731 		packet_increment_head(&po->tx_ring);
2732 		len_sum += tp_len;
2733 	} while (likely((ph != NULL) ||
2734 		/* Note: packet_read_pending() might be slow if we have
2735 		 * to call it as it's per_cpu variable, but in fast-path
2736 		 * we already short-circuit the loop with the first
2737 		 * condition, and luckily don't have to go that path
2738 		 * anyway.
2739 		 */
2740 		 (need_wait && packet_read_pending(&po->tx_ring))));
2741 
2742 	err = len_sum;
2743 	goto out_put;
2744 
2745 out_status:
2746 	__packet_set_status(po, ph, status);
2747 	kfree_skb(skb);
2748 out_put:
2749 	dev_put(dev);
2750 out:
2751 	mutex_unlock(&po->pg_vec_lock);
2752 	return err;
2753 }
2754 
2755 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2756 				        size_t reserve, size_t len,
2757 				        size_t linear, int noblock,
2758 				        int *err)
2759 {
2760 	struct sk_buff *skb;
2761 
2762 	/* Under a page?  Don't bother with paged skb. */
2763 	if (prepad + len < PAGE_SIZE || !linear)
2764 		linear = len;
2765 
2766 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2767 				   err, 0);
2768 	if (!skb)
2769 		return NULL;
2770 
2771 	skb_reserve(skb, reserve);
2772 	skb_put(skb, linear);
2773 	skb->data_len = len - linear;
2774 	skb->len += len - linear;
2775 
2776 	return skb;
2777 }
2778 
2779 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2780 {
2781 	struct sock *sk = sock->sk;
2782 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2783 	struct sk_buff *skb;
2784 	struct net_device *dev;
2785 	__be16 proto;
2786 	unsigned char *addr;
2787 	int err, reserve = 0;
2788 	struct sockcm_cookie sockc;
2789 	struct virtio_net_hdr vnet_hdr = { 0 };
2790 	int offset = 0;
2791 	struct packet_sock *po = pkt_sk(sk);
2792 	int hlen, tlen, linear;
2793 	int extra_len = 0;
2794 
2795 	/*
2796 	 *	Get and verify the address.
2797 	 */
2798 
2799 	if (likely(saddr == NULL)) {
2800 		dev	= packet_cached_dev_get(po);
2801 		proto	= po->num;
2802 		addr	= NULL;
2803 	} else {
2804 		err = -EINVAL;
2805 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2806 			goto out;
2807 		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2808 			goto out;
2809 		proto	= saddr->sll_protocol;
2810 		addr	= saddr->sll_addr;
2811 		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2812 	}
2813 
2814 	err = -ENXIO;
2815 	if (unlikely(dev == NULL))
2816 		goto out_unlock;
2817 	err = -ENETDOWN;
2818 	if (unlikely(!(dev->flags & IFF_UP)))
2819 		goto out_unlock;
2820 
2821 	sockc.tsflags = sk->sk_tsflags;
2822 	sockc.mark = sk->sk_mark;
2823 	if (msg->msg_controllen) {
2824 		err = sock_cmsg_send(sk, msg, &sockc);
2825 		if (unlikely(err))
2826 			goto out_unlock;
2827 	}
2828 
2829 	if (sock->type == SOCK_RAW)
2830 		reserve = dev->hard_header_len;
2831 	if (po->has_vnet_hdr) {
2832 		err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2833 		if (err)
2834 			goto out_unlock;
2835 	}
2836 
2837 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2838 		if (!netif_supports_nofcs(dev)) {
2839 			err = -EPROTONOSUPPORT;
2840 			goto out_unlock;
2841 		}
2842 		extra_len = 4; /* We're doing our own CRC */
2843 	}
2844 
2845 	err = -EMSGSIZE;
2846 	if (!vnet_hdr.gso_type &&
2847 	    (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2848 		goto out_unlock;
2849 
2850 	err = -ENOBUFS;
2851 	hlen = LL_RESERVED_SPACE(dev);
2852 	tlen = dev->needed_tailroom;
2853 	linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2854 	linear = max(linear, min_t(int, len, dev->hard_header_len));
2855 	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2856 			       msg->msg_flags & MSG_DONTWAIT, &err);
2857 	if (skb == NULL)
2858 		goto out_unlock;
2859 
2860 	skb_set_network_header(skb, reserve);
2861 
2862 	err = -EINVAL;
2863 	if (sock->type == SOCK_DGRAM) {
2864 		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2865 		if (unlikely(offset < 0))
2866 			goto out_free;
2867 	}
2868 
2869 	/* Returns -EFAULT on error */
2870 	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2871 	if (err)
2872 		goto out_free;
2873 
2874 	if (sock->type == SOCK_RAW &&
2875 	    !dev_validate_header(dev, skb->data, len)) {
2876 		err = -EINVAL;
2877 		goto out_free;
2878 	}
2879 
2880 	sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2881 
2882 	if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2883 	    !packet_extra_vlan_len_allowed(dev, skb)) {
2884 		err = -EMSGSIZE;
2885 		goto out_free;
2886 	}
2887 
2888 	skb->protocol = proto;
2889 	skb->dev = dev;
2890 	skb->priority = sk->sk_priority;
2891 	skb->mark = sockc.mark;
2892 
2893 	packet_pick_tx_queue(dev, skb);
2894 
2895 	if (po->has_vnet_hdr) {
2896 		err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2897 		if (err)
2898 			goto out_free;
2899 		len += sizeof(vnet_hdr);
2900 	}
2901 
2902 	skb_probe_transport_header(skb, reserve);
2903 
2904 	if (unlikely(extra_len == 4))
2905 		skb->no_fcs = 1;
2906 
2907 	err = po->xmit(skb);
2908 	if (err > 0 && (err = net_xmit_errno(err)) != 0)
2909 		goto out_unlock;
2910 
2911 	dev_put(dev);
2912 
2913 	return len;
2914 
2915 out_free:
2916 	kfree_skb(skb);
2917 out_unlock:
2918 	if (dev)
2919 		dev_put(dev);
2920 out:
2921 	return err;
2922 }
2923 
2924 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2925 {
2926 	struct sock *sk = sock->sk;
2927 	struct packet_sock *po = pkt_sk(sk);
2928 
2929 	if (po->tx_ring.pg_vec)
2930 		return tpacket_snd(po, msg);
2931 	else
2932 		return packet_snd(sock, msg, len);
2933 }
2934 
2935 /*
2936  *	Close a PACKET socket. This is fairly simple. We immediately go
2937  *	to 'closed' state and remove our protocol entry in the device list.
2938  */
2939 
2940 static int packet_release(struct socket *sock)
2941 {
2942 	struct sock *sk = sock->sk;
2943 	struct packet_sock *po;
2944 	struct packet_fanout *f;
2945 	struct net *net;
2946 	union tpacket_req_u req_u;
2947 
2948 	if (!sk)
2949 		return 0;
2950 
2951 	net = sock_net(sk);
2952 	po = pkt_sk(sk);
2953 
2954 	mutex_lock(&net->packet.sklist_lock);
2955 	sk_del_node_init_rcu(sk);
2956 	mutex_unlock(&net->packet.sklist_lock);
2957 
2958 	preempt_disable();
2959 	sock_prot_inuse_add(net, sk->sk_prot, -1);
2960 	preempt_enable();
2961 
2962 	spin_lock(&po->bind_lock);
2963 	unregister_prot_hook(sk, false);
2964 	packet_cached_dev_reset(po);
2965 
2966 	if (po->prot_hook.dev) {
2967 		dev_put(po->prot_hook.dev);
2968 		po->prot_hook.dev = NULL;
2969 	}
2970 	spin_unlock(&po->bind_lock);
2971 
2972 	packet_flush_mclist(sk);
2973 
2974 	if (po->rx_ring.pg_vec) {
2975 		memset(&req_u, 0, sizeof(req_u));
2976 		packet_set_ring(sk, &req_u, 1, 0);
2977 	}
2978 
2979 	if (po->tx_ring.pg_vec) {
2980 		memset(&req_u, 0, sizeof(req_u));
2981 		packet_set_ring(sk, &req_u, 1, 1);
2982 	}
2983 
2984 	f = fanout_release(sk);
2985 
2986 	synchronize_net();
2987 
2988 	if (f) {
2989 		fanout_release_data(f);
2990 		kfree(f);
2991 	}
2992 	/*
2993 	 *	Now the socket is dead. No more input will appear.
2994 	 */
2995 	sock_orphan(sk);
2996 	sock->sk = NULL;
2997 
2998 	/* Purge queues */
2999 
3000 	skb_queue_purge(&sk->sk_receive_queue);
3001 	packet_free_pending(po);
3002 	sk_refcnt_debug_release(sk);
3003 
3004 	sock_put(sk);
3005 	return 0;
3006 }
3007 
3008 /*
3009  *	Attach a packet hook.
3010  */
3011 
3012 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3013 			  __be16 proto)
3014 {
3015 	struct packet_sock *po = pkt_sk(sk);
3016 	struct net_device *dev_curr;
3017 	__be16 proto_curr;
3018 	bool need_rehook;
3019 	struct net_device *dev = NULL;
3020 	int ret = 0;
3021 	bool unlisted = false;
3022 
3023 	if (po->fanout)
3024 		return -EINVAL;
3025 
3026 	lock_sock(sk);
3027 	spin_lock(&po->bind_lock);
3028 	rcu_read_lock();
3029 
3030 	if (name) {
3031 		dev = dev_get_by_name_rcu(sock_net(sk), name);
3032 		if (!dev) {
3033 			ret = -ENODEV;
3034 			goto out_unlock;
3035 		}
3036 	} else if (ifindex) {
3037 		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3038 		if (!dev) {
3039 			ret = -ENODEV;
3040 			goto out_unlock;
3041 		}
3042 	}
3043 
3044 	if (dev)
3045 		dev_hold(dev);
3046 
3047 	proto_curr = po->prot_hook.type;
3048 	dev_curr = po->prot_hook.dev;
3049 
3050 	need_rehook = proto_curr != proto || dev_curr != dev;
3051 
3052 	if (need_rehook) {
3053 		if (po->running) {
3054 			rcu_read_unlock();
3055 			__unregister_prot_hook(sk, true);
3056 			rcu_read_lock();
3057 			dev_curr = po->prot_hook.dev;
3058 			if (dev)
3059 				unlisted = !dev_get_by_index_rcu(sock_net(sk),
3060 								 dev->ifindex);
3061 		}
3062 
3063 		po->num = proto;
3064 		po->prot_hook.type = proto;
3065 
3066 		if (unlikely(unlisted)) {
3067 			dev_put(dev);
3068 			po->prot_hook.dev = NULL;
3069 			po->ifindex = -1;
3070 			packet_cached_dev_reset(po);
3071 		} else {
3072 			po->prot_hook.dev = dev;
3073 			po->ifindex = dev ? dev->ifindex : 0;
3074 			packet_cached_dev_assign(po, dev);
3075 		}
3076 	}
3077 	if (dev_curr)
3078 		dev_put(dev_curr);
3079 
3080 	if (proto == 0 || !need_rehook)
3081 		goto out_unlock;
3082 
3083 	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3084 		register_prot_hook(sk);
3085 	} else {
3086 		sk->sk_err = ENETDOWN;
3087 		if (!sock_flag(sk, SOCK_DEAD))
3088 			sk->sk_error_report(sk);
3089 	}
3090 
3091 out_unlock:
3092 	rcu_read_unlock();
3093 	spin_unlock(&po->bind_lock);
3094 	release_sock(sk);
3095 	return ret;
3096 }
3097 
3098 /*
3099  *	Bind a packet socket to a device
3100  */
3101 
3102 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3103 			    int addr_len)
3104 {
3105 	struct sock *sk = sock->sk;
3106 	char name[sizeof(uaddr->sa_data) + 1];
3107 
3108 	/*
3109 	 *	Check legality
3110 	 */
3111 
3112 	if (addr_len != sizeof(struct sockaddr))
3113 		return -EINVAL;
3114 	/* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3115 	 * zero-terminated.
3116 	 */
3117 	memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3118 	name[sizeof(uaddr->sa_data)] = 0;
3119 
3120 	return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3121 }
3122 
3123 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3124 {
3125 	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3126 	struct sock *sk = sock->sk;
3127 
3128 	/*
3129 	 *	Check legality
3130 	 */
3131 
3132 	if (addr_len < sizeof(struct sockaddr_ll))
3133 		return -EINVAL;
3134 	if (sll->sll_family != AF_PACKET)
3135 		return -EINVAL;
3136 
3137 	return packet_do_bind(sk, NULL, sll->sll_ifindex,
3138 			      sll->sll_protocol ? : pkt_sk(sk)->num);
3139 }
3140 
3141 static struct proto packet_proto = {
3142 	.name	  = "PACKET",
3143 	.owner	  = THIS_MODULE,
3144 	.obj_size = sizeof(struct packet_sock),
3145 };
3146 
3147 /*
3148  *	Create a packet of type SOCK_PACKET.
3149  */
3150 
3151 static int packet_create(struct net *net, struct socket *sock, int protocol,
3152 			 int kern)
3153 {
3154 	struct sock *sk;
3155 	struct packet_sock *po;
3156 	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3157 	int err;
3158 
3159 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3160 		return -EPERM;
3161 	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3162 	    sock->type != SOCK_PACKET)
3163 		return -ESOCKTNOSUPPORT;
3164 
3165 	sock->state = SS_UNCONNECTED;
3166 
3167 	err = -ENOBUFS;
3168 	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3169 	if (sk == NULL)
3170 		goto out;
3171 
3172 	sock->ops = &packet_ops;
3173 	if (sock->type == SOCK_PACKET)
3174 		sock->ops = &packet_ops_spkt;
3175 
3176 	sock_init_data(sock, sk);
3177 
3178 	po = pkt_sk(sk);
3179 	sk->sk_family = PF_PACKET;
3180 	po->num = proto;
3181 	po->xmit = dev_queue_xmit;
3182 
3183 	err = packet_alloc_pending(po);
3184 	if (err)
3185 		goto out2;
3186 
3187 	packet_cached_dev_reset(po);
3188 
3189 	sk->sk_destruct = packet_sock_destruct;
3190 	sk_refcnt_debug_inc(sk);
3191 
3192 	/*
3193 	 *	Attach a protocol block
3194 	 */
3195 
3196 	spin_lock_init(&po->bind_lock);
3197 	mutex_init(&po->pg_vec_lock);
3198 	po->rollover = NULL;
3199 	po->prot_hook.func = packet_rcv;
3200 
3201 	if (sock->type == SOCK_PACKET)
3202 		po->prot_hook.func = packet_rcv_spkt;
3203 
3204 	po->prot_hook.af_packet_priv = sk;
3205 
3206 	if (proto) {
3207 		po->prot_hook.type = proto;
3208 		register_prot_hook(sk);
3209 	}
3210 
3211 	mutex_lock(&net->packet.sklist_lock);
3212 	sk_add_node_rcu(sk, &net->packet.sklist);
3213 	mutex_unlock(&net->packet.sklist_lock);
3214 
3215 	preempt_disable();
3216 	sock_prot_inuse_add(net, &packet_proto, 1);
3217 	preempt_enable();
3218 
3219 	return 0;
3220 out2:
3221 	sk_free(sk);
3222 out:
3223 	return err;
3224 }
3225 
3226 /*
3227  *	Pull a packet from our receive queue and hand it to the user.
3228  *	If necessary we block.
3229  */
3230 
3231 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3232 			  int flags)
3233 {
3234 	struct sock *sk = sock->sk;
3235 	struct sk_buff *skb;
3236 	int copied, err;
3237 	int vnet_hdr_len = 0;
3238 	unsigned int origlen = 0;
3239 
3240 	err = -EINVAL;
3241 	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3242 		goto out;
3243 
3244 #if 0
3245 	/* What error should we return now? EUNATTACH? */
3246 	if (pkt_sk(sk)->ifindex < 0)
3247 		return -ENODEV;
3248 #endif
3249 
3250 	if (flags & MSG_ERRQUEUE) {
3251 		err = sock_recv_errqueue(sk, msg, len,
3252 					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3253 		goto out;
3254 	}
3255 
3256 	/*
3257 	 *	Call the generic datagram receiver. This handles all sorts
3258 	 *	of horrible races and re-entrancy so we can forget about it
3259 	 *	in the protocol layers.
3260 	 *
3261 	 *	Now it will return ENETDOWN, if device have just gone down,
3262 	 *	but then it will block.
3263 	 */
3264 
3265 	skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3266 
3267 	/*
3268 	 *	An error occurred so return it. Because skb_recv_datagram()
3269 	 *	handles the blocking we don't see and worry about blocking
3270 	 *	retries.
3271 	 */
3272 
3273 	if (skb == NULL)
3274 		goto out;
3275 
3276 	if (pkt_sk(sk)->pressure)
3277 		packet_rcv_has_room(pkt_sk(sk), NULL);
3278 
3279 	if (pkt_sk(sk)->has_vnet_hdr) {
3280 		err = packet_rcv_vnet(msg, skb, &len);
3281 		if (err)
3282 			goto out_free;
3283 		vnet_hdr_len = sizeof(struct virtio_net_hdr);
3284 	}
3285 
3286 	/* You lose any data beyond the buffer you gave. If it worries
3287 	 * a user program they can ask the device for its MTU
3288 	 * anyway.
3289 	 */
3290 	copied = skb->len;
3291 	if (copied > len) {
3292 		copied = len;
3293 		msg->msg_flags |= MSG_TRUNC;
3294 	}
3295 
3296 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3297 	if (err)
3298 		goto out_free;
3299 
3300 	if (sock->type != SOCK_PACKET) {
3301 		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3302 
3303 		/* Original length was stored in sockaddr_ll fields */
3304 		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3305 		sll->sll_family = AF_PACKET;
3306 		sll->sll_protocol = skb->protocol;
3307 	}
3308 
3309 	sock_recv_ts_and_drops(msg, sk, skb);
3310 
3311 	if (msg->msg_name) {
3312 		/* If the address length field is there to be filled
3313 		 * in, we fill it in now.
3314 		 */
3315 		if (sock->type == SOCK_PACKET) {
3316 			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3317 			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3318 		} else {
3319 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3320 
3321 			msg->msg_namelen = sll->sll_halen +
3322 				offsetof(struct sockaddr_ll, sll_addr);
3323 		}
3324 		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3325 		       msg->msg_namelen);
3326 	}
3327 
3328 	if (pkt_sk(sk)->auxdata) {
3329 		struct tpacket_auxdata aux;
3330 
3331 		aux.tp_status = TP_STATUS_USER;
3332 		if (skb->ip_summed == CHECKSUM_PARTIAL)
3333 			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3334 		else if (skb->pkt_type != PACKET_OUTGOING &&
3335 			 (skb->ip_summed == CHECKSUM_COMPLETE ||
3336 			  skb_csum_unnecessary(skb)))
3337 			aux.tp_status |= TP_STATUS_CSUM_VALID;
3338 
3339 		aux.tp_len = origlen;
3340 		aux.tp_snaplen = skb->len;
3341 		aux.tp_mac = 0;
3342 		aux.tp_net = skb_network_offset(skb);
3343 		if (skb_vlan_tag_present(skb)) {
3344 			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3345 			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3346 			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3347 		} else {
3348 			aux.tp_vlan_tci = 0;
3349 			aux.tp_vlan_tpid = 0;
3350 		}
3351 		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3352 	}
3353 
3354 	/*
3355 	 *	Free or return the buffer as appropriate. Again this
3356 	 *	hides all the races and re-entrancy issues from us.
3357 	 */
3358 	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3359 
3360 out_free:
3361 	skb_free_datagram(sk, skb);
3362 out:
3363 	return err;
3364 }
3365 
3366 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3367 			       int *uaddr_len, int peer)
3368 {
3369 	struct net_device *dev;
3370 	struct sock *sk	= sock->sk;
3371 
3372 	if (peer)
3373 		return -EOPNOTSUPP;
3374 
3375 	uaddr->sa_family = AF_PACKET;
3376 	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3377 	rcu_read_lock();
3378 	dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3379 	if (dev)
3380 		strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3381 	rcu_read_unlock();
3382 	*uaddr_len = sizeof(*uaddr);
3383 
3384 	return 0;
3385 }
3386 
3387 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3388 			  int *uaddr_len, int peer)
3389 {
3390 	struct net_device *dev;
3391 	struct sock *sk = sock->sk;
3392 	struct packet_sock *po = pkt_sk(sk);
3393 	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3394 
3395 	if (peer)
3396 		return -EOPNOTSUPP;
3397 
3398 	sll->sll_family = AF_PACKET;
3399 	sll->sll_ifindex = po->ifindex;
3400 	sll->sll_protocol = po->num;
3401 	sll->sll_pkttype = 0;
3402 	rcu_read_lock();
3403 	dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3404 	if (dev) {
3405 		sll->sll_hatype = dev->type;
3406 		sll->sll_halen = dev->addr_len;
3407 		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3408 	} else {
3409 		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3410 		sll->sll_halen = 0;
3411 	}
3412 	rcu_read_unlock();
3413 	*uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3414 
3415 	return 0;
3416 }
3417 
3418 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3419 			 int what)
3420 {
3421 	switch (i->type) {
3422 	case PACKET_MR_MULTICAST:
3423 		if (i->alen != dev->addr_len)
3424 			return -EINVAL;
3425 		if (what > 0)
3426 			return dev_mc_add(dev, i->addr);
3427 		else
3428 			return dev_mc_del(dev, i->addr);
3429 		break;
3430 	case PACKET_MR_PROMISC:
3431 		return dev_set_promiscuity(dev, what);
3432 	case PACKET_MR_ALLMULTI:
3433 		return dev_set_allmulti(dev, what);
3434 	case PACKET_MR_UNICAST:
3435 		if (i->alen != dev->addr_len)
3436 			return -EINVAL;
3437 		if (what > 0)
3438 			return dev_uc_add(dev, i->addr);
3439 		else
3440 			return dev_uc_del(dev, i->addr);
3441 		break;
3442 	default:
3443 		break;
3444 	}
3445 	return 0;
3446 }
3447 
3448 static void packet_dev_mclist_delete(struct net_device *dev,
3449 				     struct packet_mclist **mlp)
3450 {
3451 	struct packet_mclist *ml;
3452 
3453 	while ((ml = *mlp) != NULL) {
3454 		if (ml->ifindex == dev->ifindex) {
3455 			packet_dev_mc(dev, ml, -1);
3456 			*mlp = ml->next;
3457 			kfree(ml);
3458 		} else
3459 			mlp = &ml->next;
3460 	}
3461 }
3462 
3463 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3464 {
3465 	struct packet_sock *po = pkt_sk(sk);
3466 	struct packet_mclist *ml, *i;
3467 	struct net_device *dev;
3468 	int err;
3469 
3470 	rtnl_lock();
3471 
3472 	err = -ENODEV;
3473 	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3474 	if (!dev)
3475 		goto done;
3476 
3477 	err = -EINVAL;
3478 	if (mreq->mr_alen > dev->addr_len)
3479 		goto done;
3480 
3481 	err = -ENOBUFS;
3482 	i = kmalloc(sizeof(*i), GFP_KERNEL);
3483 	if (i == NULL)
3484 		goto done;
3485 
3486 	err = 0;
3487 	for (ml = po->mclist; ml; ml = ml->next) {
3488 		if (ml->ifindex == mreq->mr_ifindex &&
3489 		    ml->type == mreq->mr_type &&
3490 		    ml->alen == mreq->mr_alen &&
3491 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3492 			ml->count++;
3493 			/* Free the new element ... */
3494 			kfree(i);
3495 			goto done;
3496 		}
3497 	}
3498 
3499 	i->type = mreq->mr_type;
3500 	i->ifindex = mreq->mr_ifindex;
3501 	i->alen = mreq->mr_alen;
3502 	memcpy(i->addr, mreq->mr_address, i->alen);
3503 	memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3504 	i->count = 1;
3505 	i->next = po->mclist;
3506 	po->mclist = i;
3507 	err = packet_dev_mc(dev, i, 1);
3508 	if (err) {
3509 		po->mclist = i->next;
3510 		kfree(i);
3511 	}
3512 
3513 done:
3514 	rtnl_unlock();
3515 	return err;
3516 }
3517 
3518 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3519 {
3520 	struct packet_mclist *ml, **mlp;
3521 
3522 	rtnl_lock();
3523 
3524 	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3525 		if (ml->ifindex == mreq->mr_ifindex &&
3526 		    ml->type == mreq->mr_type &&
3527 		    ml->alen == mreq->mr_alen &&
3528 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3529 			if (--ml->count == 0) {
3530 				struct net_device *dev;
3531 				*mlp = ml->next;
3532 				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3533 				if (dev)
3534 					packet_dev_mc(dev, ml, -1);
3535 				kfree(ml);
3536 			}
3537 			break;
3538 		}
3539 	}
3540 	rtnl_unlock();
3541 	return 0;
3542 }
3543 
3544 static void packet_flush_mclist(struct sock *sk)
3545 {
3546 	struct packet_sock *po = pkt_sk(sk);
3547 	struct packet_mclist *ml;
3548 
3549 	if (!po->mclist)
3550 		return;
3551 
3552 	rtnl_lock();
3553 	while ((ml = po->mclist) != NULL) {
3554 		struct net_device *dev;
3555 
3556 		po->mclist = ml->next;
3557 		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3558 		if (dev != NULL)
3559 			packet_dev_mc(dev, ml, -1);
3560 		kfree(ml);
3561 	}
3562 	rtnl_unlock();
3563 }
3564 
3565 static int
3566 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3567 {
3568 	struct sock *sk = sock->sk;
3569 	struct packet_sock *po = pkt_sk(sk);
3570 	int ret;
3571 
3572 	if (level != SOL_PACKET)
3573 		return -ENOPROTOOPT;
3574 
3575 	switch (optname) {
3576 	case PACKET_ADD_MEMBERSHIP:
3577 	case PACKET_DROP_MEMBERSHIP:
3578 	{
3579 		struct packet_mreq_max mreq;
3580 		int len = optlen;
3581 		memset(&mreq, 0, sizeof(mreq));
3582 		if (len < sizeof(struct packet_mreq))
3583 			return -EINVAL;
3584 		if (len > sizeof(mreq))
3585 			len = sizeof(mreq);
3586 		if (copy_from_user(&mreq, optval, len))
3587 			return -EFAULT;
3588 		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3589 			return -EINVAL;
3590 		if (optname == PACKET_ADD_MEMBERSHIP)
3591 			ret = packet_mc_add(sk, &mreq);
3592 		else
3593 			ret = packet_mc_drop(sk, &mreq);
3594 		return ret;
3595 	}
3596 
3597 	case PACKET_RX_RING:
3598 	case PACKET_TX_RING:
3599 	{
3600 		union tpacket_req_u req_u;
3601 		int len;
3602 
3603 		switch (po->tp_version) {
3604 		case TPACKET_V1:
3605 		case TPACKET_V2:
3606 			len = sizeof(req_u.req);
3607 			break;
3608 		case TPACKET_V3:
3609 		default:
3610 			len = sizeof(req_u.req3);
3611 			break;
3612 		}
3613 		if (optlen < len)
3614 			return -EINVAL;
3615 		if (copy_from_user(&req_u.req, optval, len))
3616 			return -EFAULT;
3617 		return packet_set_ring(sk, &req_u, 0,
3618 			optname == PACKET_TX_RING);
3619 	}
3620 	case PACKET_COPY_THRESH:
3621 	{
3622 		int val;
3623 
3624 		if (optlen != sizeof(val))
3625 			return -EINVAL;
3626 		if (copy_from_user(&val, optval, sizeof(val)))
3627 			return -EFAULT;
3628 
3629 		pkt_sk(sk)->copy_thresh = val;
3630 		return 0;
3631 	}
3632 	case PACKET_VERSION:
3633 	{
3634 		int val;
3635 
3636 		if (optlen != sizeof(val))
3637 			return -EINVAL;
3638 		if (copy_from_user(&val, optval, sizeof(val)))
3639 			return -EFAULT;
3640 		switch (val) {
3641 		case TPACKET_V1:
3642 		case TPACKET_V2:
3643 		case TPACKET_V3:
3644 			break;
3645 		default:
3646 			return -EINVAL;
3647 		}
3648 		lock_sock(sk);
3649 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3650 			ret = -EBUSY;
3651 		} else {
3652 			po->tp_version = val;
3653 			ret = 0;
3654 		}
3655 		release_sock(sk);
3656 		return ret;
3657 	}
3658 	case PACKET_RESERVE:
3659 	{
3660 		unsigned int val;
3661 
3662 		if (optlen != sizeof(val))
3663 			return -EINVAL;
3664 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3665 			return -EBUSY;
3666 		if (copy_from_user(&val, optval, sizeof(val)))
3667 			return -EFAULT;
3668 		po->tp_reserve = val;
3669 		return 0;
3670 	}
3671 	case PACKET_LOSS:
3672 	{
3673 		unsigned int val;
3674 
3675 		if (optlen != sizeof(val))
3676 			return -EINVAL;
3677 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3678 			return -EBUSY;
3679 		if (copy_from_user(&val, optval, sizeof(val)))
3680 			return -EFAULT;
3681 		po->tp_loss = !!val;
3682 		return 0;
3683 	}
3684 	case PACKET_AUXDATA:
3685 	{
3686 		int val;
3687 
3688 		if (optlen < sizeof(val))
3689 			return -EINVAL;
3690 		if (copy_from_user(&val, optval, sizeof(val)))
3691 			return -EFAULT;
3692 
3693 		po->auxdata = !!val;
3694 		return 0;
3695 	}
3696 	case PACKET_ORIGDEV:
3697 	{
3698 		int val;
3699 
3700 		if (optlen < sizeof(val))
3701 			return -EINVAL;
3702 		if (copy_from_user(&val, optval, sizeof(val)))
3703 			return -EFAULT;
3704 
3705 		po->origdev = !!val;
3706 		return 0;
3707 	}
3708 	case PACKET_VNET_HDR:
3709 	{
3710 		int val;
3711 
3712 		if (sock->type != SOCK_RAW)
3713 			return -EINVAL;
3714 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3715 			return -EBUSY;
3716 		if (optlen < sizeof(val))
3717 			return -EINVAL;
3718 		if (copy_from_user(&val, optval, sizeof(val)))
3719 			return -EFAULT;
3720 
3721 		po->has_vnet_hdr = !!val;
3722 		return 0;
3723 	}
3724 	case PACKET_TIMESTAMP:
3725 	{
3726 		int val;
3727 
3728 		if (optlen != sizeof(val))
3729 			return -EINVAL;
3730 		if (copy_from_user(&val, optval, sizeof(val)))
3731 			return -EFAULT;
3732 
3733 		po->tp_tstamp = val;
3734 		return 0;
3735 	}
3736 	case PACKET_FANOUT:
3737 	{
3738 		int val;
3739 
3740 		if (optlen != sizeof(val))
3741 			return -EINVAL;
3742 		if (copy_from_user(&val, optval, sizeof(val)))
3743 			return -EFAULT;
3744 
3745 		return fanout_add(sk, val & 0xffff, val >> 16);
3746 	}
3747 	case PACKET_FANOUT_DATA:
3748 	{
3749 		if (!po->fanout)
3750 			return -EINVAL;
3751 
3752 		return fanout_set_data(po, optval, optlen);
3753 	}
3754 	case PACKET_TX_HAS_OFF:
3755 	{
3756 		unsigned int val;
3757 
3758 		if (optlen != sizeof(val))
3759 			return -EINVAL;
3760 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3761 			return -EBUSY;
3762 		if (copy_from_user(&val, optval, sizeof(val)))
3763 			return -EFAULT;
3764 		po->tp_tx_has_off = !!val;
3765 		return 0;
3766 	}
3767 	case PACKET_QDISC_BYPASS:
3768 	{
3769 		int val;
3770 
3771 		if (optlen != sizeof(val))
3772 			return -EINVAL;
3773 		if (copy_from_user(&val, optval, sizeof(val)))
3774 			return -EFAULT;
3775 
3776 		po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3777 		return 0;
3778 	}
3779 	default:
3780 		return -ENOPROTOOPT;
3781 	}
3782 }
3783 
3784 static int packet_getsockopt(struct socket *sock, int level, int optname,
3785 			     char __user *optval, int __user *optlen)
3786 {
3787 	int len;
3788 	int val, lv = sizeof(val);
3789 	struct sock *sk = sock->sk;
3790 	struct packet_sock *po = pkt_sk(sk);
3791 	void *data = &val;
3792 	union tpacket_stats_u st;
3793 	struct tpacket_rollover_stats rstats;
3794 
3795 	if (level != SOL_PACKET)
3796 		return -ENOPROTOOPT;
3797 
3798 	if (get_user(len, optlen))
3799 		return -EFAULT;
3800 
3801 	if (len < 0)
3802 		return -EINVAL;
3803 
3804 	switch (optname) {
3805 	case PACKET_STATISTICS:
3806 		spin_lock_bh(&sk->sk_receive_queue.lock);
3807 		memcpy(&st, &po->stats, sizeof(st));
3808 		memset(&po->stats, 0, sizeof(po->stats));
3809 		spin_unlock_bh(&sk->sk_receive_queue.lock);
3810 
3811 		if (po->tp_version == TPACKET_V3) {
3812 			lv = sizeof(struct tpacket_stats_v3);
3813 			st.stats3.tp_packets += st.stats3.tp_drops;
3814 			data = &st.stats3;
3815 		} else {
3816 			lv = sizeof(struct tpacket_stats);
3817 			st.stats1.tp_packets += st.stats1.tp_drops;
3818 			data = &st.stats1;
3819 		}
3820 
3821 		break;
3822 	case PACKET_AUXDATA:
3823 		val = po->auxdata;
3824 		break;
3825 	case PACKET_ORIGDEV:
3826 		val = po->origdev;
3827 		break;
3828 	case PACKET_VNET_HDR:
3829 		val = po->has_vnet_hdr;
3830 		break;
3831 	case PACKET_VERSION:
3832 		val = po->tp_version;
3833 		break;
3834 	case PACKET_HDRLEN:
3835 		if (len > sizeof(int))
3836 			len = sizeof(int);
3837 		if (copy_from_user(&val, optval, len))
3838 			return -EFAULT;
3839 		switch (val) {
3840 		case TPACKET_V1:
3841 			val = sizeof(struct tpacket_hdr);
3842 			break;
3843 		case TPACKET_V2:
3844 			val = sizeof(struct tpacket2_hdr);
3845 			break;
3846 		case TPACKET_V3:
3847 			val = sizeof(struct tpacket3_hdr);
3848 			break;
3849 		default:
3850 			return -EINVAL;
3851 		}
3852 		break;
3853 	case PACKET_RESERVE:
3854 		val = po->tp_reserve;
3855 		break;
3856 	case PACKET_LOSS:
3857 		val = po->tp_loss;
3858 		break;
3859 	case PACKET_TIMESTAMP:
3860 		val = po->tp_tstamp;
3861 		break;
3862 	case PACKET_FANOUT:
3863 		val = (po->fanout ?
3864 		       ((u32)po->fanout->id |
3865 			((u32)po->fanout->type << 16) |
3866 			((u32)po->fanout->flags << 24)) :
3867 		       0);
3868 		break;
3869 	case PACKET_ROLLOVER_STATS:
3870 		if (!po->rollover)
3871 			return -EINVAL;
3872 		rstats.tp_all = atomic_long_read(&po->rollover->num);
3873 		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3874 		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3875 		data = &rstats;
3876 		lv = sizeof(rstats);
3877 		break;
3878 	case PACKET_TX_HAS_OFF:
3879 		val = po->tp_tx_has_off;
3880 		break;
3881 	case PACKET_QDISC_BYPASS:
3882 		val = packet_use_direct_xmit(po);
3883 		break;
3884 	default:
3885 		return -ENOPROTOOPT;
3886 	}
3887 
3888 	if (len > lv)
3889 		len = lv;
3890 	if (put_user(len, optlen))
3891 		return -EFAULT;
3892 	if (copy_to_user(optval, data, len))
3893 		return -EFAULT;
3894 	return 0;
3895 }
3896 
3897 
3898 #ifdef CONFIG_COMPAT
3899 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3900 				    char __user *optval, unsigned int optlen)
3901 {
3902 	struct packet_sock *po = pkt_sk(sock->sk);
3903 
3904 	if (level != SOL_PACKET)
3905 		return -ENOPROTOOPT;
3906 
3907 	if (optname == PACKET_FANOUT_DATA &&
3908 	    po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3909 		optval = (char __user *)get_compat_bpf_fprog(optval);
3910 		if (!optval)
3911 			return -EFAULT;
3912 		optlen = sizeof(struct sock_fprog);
3913 	}
3914 
3915 	return packet_setsockopt(sock, level, optname, optval, optlen);
3916 }
3917 #endif
3918 
3919 static int packet_notifier(struct notifier_block *this,
3920 			   unsigned long msg, void *ptr)
3921 {
3922 	struct sock *sk;
3923 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3924 	struct net *net = dev_net(dev);
3925 
3926 	rcu_read_lock();
3927 	sk_for_each_rcu(sk, &net->packet.sklist) {
3928 		struct packet_sock *po = pkt_sk(sk);
3929 
3930 		switch (msg) {
3931 		case NETDEV_UNREGISTER:
3932 			if (po->mclist)
3933 				packet_dev_mclist_delete(dev, &po->mclist);
3934 			/* fallthrough */
3935 
3936 		case NETDEV_DOWN:
3937 			if (dev->ifindex == po->ifindex) {
3938 				spin_lock(&po->bind_lock);
3939 				if (po->running) {
3940 					__unregister_prot_hook(sk, false);
3941 					sk->sk_err = ENETDOWN;
3942 					if (!sock_flag(sk, SOCK_DEAD))
3943 						sk->sk_error_report(sk);
3944 				}
3945 				if (msg == NETDEV_UNREGISTER) {
3946 					packet_cached_dev_reset(po);
3947 					po->ifindex = -1;
3948 					if (po->prot_hook.dev)
3949 						dev_put(po->prot_hook.dev);
3950 					po->prot_hook.dev = NULL;
3951 				}
3952 				spin_unlock(&po->bind_lock);
3953 			}
3954 			break;
3955 		case NETDEV_UP:
3956 			if (dev->ifindex == po->ifindex) {
3957 				spin_lock(&po->bind_lock);
3958 				if (po->num)
3959 					register_prot_hook(sk);
3960 				spin_unlock(&po->bind_lock);
3961 			}
3962 			break;
3963 		}
3964 	}
3965 	rcu_read_unlock();
3966 	return NOTIFY_DONE;
3967 }
3968 
3969 
3970 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3971 			unsigned long arg)
3972 {
3973 	struct sock *sk = sock->sk;
3974 
3975 	switch (cmd) {
3976 	case SIOCOUTQ:
3977 	{
3978 		int amount = sk_wmem_alloc_get(sk);
3979 
3980 		return put_user(amount, (int __user *)arg);
3981 	}
3982 	case SIOCINQ:
3983 	{
3984 		struct sk_buff *skb;
3985 		int amount = 0;
3986 
3987 		spin_lock_bh(&sk->sk_receive_queue.lock);
3988 		skb = skb_peek(&sk->sk_receive_queue);
3989 		if (skb)
3990 			amount = skb->len;
3991 		spin_unlock_bh(&sk->sk_receive_queue.lock);
3992 		return put_user(amount, (int __user *)arg);
3993 	}
3994 	case SIOCGSTAMP:
3995 		return sock_get_timestamp(sk, (struct timeval __user *)arg);
3996 	case SIOCGSTAMPNS:
3997 		return sock_get_timestampns(sk, (struct timespec __user *)arg);
3998 
3999 #ifdef CONFIG_INET
4000 	case SIOCADDRT:
4001 	case SIOCDELRT:
4002 	case SIOCDARP:
4003 	case SIOCGARP:
4004 	case SIOCSARP:
4005 	case SIOCGIFADDR:
4006 	case SIOCSIFADDR:
4007 	case SIOCGIFBRDADDR:
4008 	case SIOCSIFBRDADDR:
4009 	case SIOCGIFNETMASK:
4010 	case SIOCSIFNETMASK:
4011 	case SIOCGIFDSTADDR:
4012 	case SIOCSIFDSTADDR:
4013 	case SIOCSIFFLAGS:
4014 		return inet_dgram_ops.ioctl(sock, cmd, arg);
4015 #endif
4016 
4017 	default:
4018 		return -ENOIOCTLCMD;
4019 	}
4020 	return 0;
4021 }
4022 
4023 static unsigned int packet_poll(struct file *file, struct socket *sock,
4024 				poll_table *wait)
4025 {
4026 	struct sock *sk = sock->sk;
4027 	struct packet_sock *po = pkt_sk(sk);
4028 	unsigned int mask = datagram_poll(file, sock, wait);
4029 
4030 	spin_lock_bh(&sk->sk_receive_queue.lock);
4031 	if (po->rx_ring.pg_vec) {
4032 		if (!packet_previous_rx_frame(po, &po->rx_ring,
4033 			TP_STATUS_KERNEL))
4034 			mask |= POLLIN | POLLRDNORM;
4035 	}
4036 	if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4037 		po->pressure = 0;
4038 	spin_unlock_bh(&sk->sk_receive_queue.lock);
4039 	spin_lock_bh(&sk->sk_write_queue.lock);
4040 	if (po->tx_ring.pg_vec) {
4041 		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4042 			mask |= POLLOUT | POLLWRNORM;
4043 	}
4044 	spin_unlock_bh(&sk->sk_write_queue.lock);
4045 	return mask;
4046 }
4047 
4048 
4049 /* Dirty? Well, I still did not learn better way to account
4050  * for user mmaps.
4051  */
4052 
4053 static void packet_mm_open(struct vm_area_struct *vma)
4054 {
4055 	struct file *file = vma->vm_file;
4056 	struct socket *sock = file->private_data;
4057 	struct sock *sk = sock->sk;
4058 
4059 	if (sk)
4060 		atomic_inc(&pkt_sk(sk)->mapped);
4061 }
4062 
4063 static void packet_mm_close(struct vm_area_struct *vma)
4064 {
4065 	struct file *file = vma->vm_file;
4066 	struct socket *sock = file->private_data;
4067 	struct sock *sk = sock->sk;
4068 
4069 	if (sk)
4070 		atomic_dec(&pkt_sk(sk)->mapped);
4071 }
4072 
4073 static const struct vm_operations_struct packet_mmap_ops = {
4074 	.open	=	packet_mm_open,
4075 	.close	=	packet_mm_close,
4076 };
4077 
4078 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4079 			unsigned int len)
4080 {
4081 	int i;
4082 
4083 	for (i = 0; i < len; i++) {
4084 		if (likely(pg_vec[i].buffer)) {
4085 			if (is_vmalloc_addr(pg_vec[i].buffer))
4086 				vfree(pg_vec[i].buffer);
4087 			else
4088 				free_pages((unsigned long)pg_vec[i].buffer,
4089 					   order);
4090 			pg_vec[i].buffer = NULL;
4091 		}
4092 	}
4093 	kfree(pg_vec);
4094 }
4095 
4096 static char *alloc_one_pg_vec_page(unsigned long order)
4097 {
4098 	char *buffer;
4099 	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4100 			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4101 
4102 	buffer = (char *) __get_free_pages(gfp_flags, order);
4103 	if (buffer)
4104 		return buffer;
4105 
4106 	/* __get_free_pages failed, fall back to vmalloc */
4107 	buffer = vzalloc((1 << order) * PAGE_SIZE);
4108 	if (buffer)
4109 		return buffer;
4110 
4111 	/* vmalloc failed, lets dig into swap here */
4112 	gfp_flags &= ~__GFP_NORETRY;
4113 	buffer = (char *) __get_free_pages(gfp_flags, order);
4114 	if (buffer)
4115 		return buffer;
4116 
4117 	/* complete and utter failure */
4118 	return NULL;
4119 }
4120 
4121 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4122 {
4123 	unsigned int block_nr = req->tp_block_nr;
4124 	struct pgv *pg_vec;
4125 	int i;
4126 
4127 	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4128 	if (unlikely(!pg_vec))
4129 		goto out;
4130 
4131 	for (i = 0; i < block_nr; i++) {
4132 		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4133 		if (unlikely(!pg_vec[i].buffer))
4134 			goto out_free_pgvec;
4135 	}
4136 
4137 out:
4138 	return pg_vec;
4139 
4140 out_free_pgvec:
4141 	free_pg_vec(pg_vec, order, block_nr);
4142 	pg_vec = NULL;
4143 	goto out;
4144 }
4145 
4146 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4147 		int closing, int tx_ring)
4148 {
4149 	struct pgv *pg_vec = NULL;
4150 	struct packet_sock *po = pkt_sk(sk);
4151 	int was_running, order = 0;
4152 	struct packet_ring_buffer *rb;
4153 	struct sk_buff_head *rb_queue;
4154 	__be16 num;
4155 	int err = -EINVAL;
4156 	/* Added to avoid minimal code churn */
4157 	struct tpacket_req *req = &req_u->req;
4158 
4159 	lock_sock(sk);
4160 
4161 	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4162 	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4163 
4164 	err = -EBUSY;
4165 	if (!closing) {
4166 		if (atomic_read(&po->mapped))
4167 			goto out;
4168 		if (packet_read_pending(rb))
4169 			goto out;
4170 	}
4171 
4172 	if (req->tp_block_nr) {
4173 		/* Sanity tests and some calculations */
4174 		err = -EBUSY;
4175 		if (unlikely(rb->pg_vec))
4176 			goto out;
4177 
4178 		switch (po->tp_version) {
4179 		case TPACKET_V1:
4180 			po->tp_hdrlen = TPACKET_HDRLEN;
4181 			break;
4182 		case TPACKET_V2:
4183 			po->tp_hdrlen = TPACKET2_HDRLEN;
4184 			break;
4185 		case TPACKET_V3:
4186 			po->tp_hdrlen = TPACKET3_HDRLEN;
4187 			break;
4188 		}
4189 
4190 		err = -EINVAL;
4191 		if (unlikely((int)req->tp_block_size <= 0))
4192 			goto out;
4193 		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4194 			goto out;
4195 		if (po->tp_version >= TPACKET_V3 &&
4196 		    (int)(req->tp_block_size -
4197 			  BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4198 			goto out;
4199 		if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4200 					po->tp_reserve))
4201 			goto out;
4202 		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4203 			goto out;
4204 
4205 		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4206 		if (unlikely(rb->frames_per_block == 0))
4207 			goto out;
4208 		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4209 					req->tp_frame_nr))
4210 			goto out;
4211 
4212 		err = -ENOMEM;
4213 		order = get_order(req->tp_block_size);
4214 		pg_vec = alloc_pg_vec(req, order);
4215 		if (unlikely(!pg_vec))
4216 			goto out;
4217 		switch (po->tp_version) {
4218 		case TPACKET_V3:
4219 			/* Block transmit is not supported yet */
4220 			if (!tx_ring) {
4221 				init_prb_bdqc(po, rb, pg_vec, req_u);
4222 			} else {
4223 				struct tpacket_req3 *req3 = &req_u->req3;
4224 
4225 				if (req3->tp_retire_blk_tov ||
4226 				    req3->tp_sizeof_priv ||
4227 				    req3->tp_feature_req_word) {
4228 					err = -EINVAL;
4229 					goto out;
4230 				}
4231 			}
4232 			break;
4233 		default:
4234 			break;
4235 		}
4236 	}
4237 	/* Done */
4238 	else {
4239 		err = -EINVAL;
4240 		if (unlikely(req->tp_frame_nr))
4241 			goto out;
4242 	}
4243 
4244 
4245 	/* Detach socket from network */
4246 	spin_lock(&po->bind_lock);
4247 	was_running = po->running;
4248 	num = po->num;
4249 	if (was_running) {
4250 		po->num = 0;
4251 		__unregister_prot_hook(sk, false);
4252 	}
4253 	spin_unlock(&po->bind_lock);
4254 
4255 	synchronize_net();
4256 
4257 	err = -EBUSY;
4258 	mutex_lock(&po->pg_vec_lock);
4259 	if (closing || atomic_read(&po->mapped) == 0) {
4260 		err = 0;
4261 		spin_lock_bh(&rb_queue->lock);
4262 		swap(rb->pg_vec, pg_vec);
4263 		rb->frame_max = (req->tp_frame_nr - 1);
4264 		rb->head = 0;
4265 		rb->frame_size = req->tp_frame_size;
4266 		spin_unlock_bh(&rb_queue->lock);
4267 
4268 		swap(rb->pg_vec_order, order);
4269 		swap(rb->pg_vec_len, req->tp_block_nr);
4270 
4271 		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4272 		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4273 						tpacket_rcv : packet_rcv;
4274 		skb_queue_purge(rb_queue);
4275 		if (atomic_read(&po->mapped))
4276 			pr_err("packet_mmap: vma is busy: %d\n",
4277 			       atomic_read(&po->mapped));
4278 	}
4279 	mutex_unlock(&po->pg_vec_lock);
4280 
4281 	spin_lock(&po->bind_lock);
4282 	if (was_running) {
4283 		po->num = num;
4284 		register_prot_hook(sk);
4285 	}
4286 	spin_unlock(&po->bind_lock);
4287 	if (closing && (po->tp_version > TPACKET_V2)) {
4288 		/* Because we don't support block-based V3 on tx-ring */
4289 		if (!tx_ring)
4290 			prb_shutdown_retire_blk_timer(po, rb_queue);
4291 	}
4292 
4293 	if (pg_vec)
4294 		free_pg_vec(pg_vec, order, req->tp_block_nr);
4295 out:
4296 	release_sock(sk);
4297 	return err;
4298 }
4299 
4300 static int packet_mmap(struct file *file, struct socket *sock,
4301 		struct vm_area_struct *vma)
4302 {
4303 	struct sock *sk = sock->sk;
4304 	struct packet_sock *po = pkt_sk(sk);
4305 	unsigned long size, expected_size;
4306 	struct packet_ring_buffer *rb;
4307 	unsigned long start;
4308 	int err = -EINVAL;
4309 	int i;
4310 
4311 	if (vma->vm_pgoff)
4312 		return -EINVAL;
4313 
4314 	mutex_lock(&po->pg_vec_lock);
4315 
4316 	expected_size = 0;
4317 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4318 		if (rb->pg_vec) {
4319 			expected_size += rb->pg_vec_len
4320 						* rb->pg_vec_pages
4321 						* PAGE_SIZE;
4322 		}
4323 	}
4324 
4325 	if (expected_size == 0)
4326 		goto out;
4327 
4328 	size = vma->vm_end - vma->vm_start;
4329 	if (size != expected_size)
4330 		goto out;
4331 
4332 	start = vma->vm_start;
4333 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4334 		if (rb->pg_vec == NULL)
4335 			continue;
4336 
4337 		for (i = 0; i < rb->pg_vec_len; i++) {
4338 			struct page *page;
4339 			void *kaddr = rb->pg_vec[i].buffer;
4340 			int pg_num;
4341 
4342 			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4343 				page = pgv_to_page(kaddr);
4344 				err = vm_insert_page(vma, start, page);
4345 				if (unlikely(err))
4346 					goto out;
4347 				start += PAGE_SIZE;
4348 				kaddr += PAGE_SIZE;
4349 			}
4350 		}
4351 	}
4352 
4353 	atomic_inc(&po->mapped);
4354 	vma->vm_ops = &packet_mmap_ops;
4355 	err = 0;
4356 
4357 out:
4358 	mutex_unlock(&po->pg_vec_lock);
4359 	return err;
4360 }
4361 
4362 static const struct proto_ops packet_ops_spkt = {
4363 	.family =	PF_PACKET,
4364 	.owner =	THIS_MODULE,
4365 	.release =	packet_release,
4366 	.bind =		packet_bind_spkt,
4367 	.connect =	sock_no_connect,
4368 	.socketpair =	sock_no_socketpair,
4369 	.accept =	sock_no_accept,
4370 	.getname =	packet_getname_spkt,
4371 	.poll =		datagram_poll,
4372 	.ioctl =	packet_ioctl,
4373 	.listen =	sock_no_listen,
4374 	.shutdown =	sock_no_shutdown,
4375 	.setsockopt =	sock_no_setsockopt,
4376 	.getsockopt =	sock_no_getsockopt,
4377 	.sendmsg =	packet_sendmsg_spkt,
4378 	.recvmsg =	packet_recvmsg,
4379 	.mmap =		sock_no_mmap,
4380 	.sendpage =	sock_no_sendpage,
4381 };
4382 
4383 static const struct proto_ops packet_ops = {
4384 	.family =	PF_PACKET,
4385 	.owner =	THIS_MODULE,
4386 	.release =	packet_release,
4387 	.bind =		packet_bind,
4388 	.connect =	sock_no_connect,
4389 	.socketpair =	sock_no_socketpair,
4390 	.accept =	sock_no_accept,
4391 	.getname =	packet_getname,
4392 	.poll =		packet_poll,
4393 	.ioctl =	packet_ioctl,
4394 	.listen =	sock_no_listen,
4395 	.shutdown =	sock_no_shutdown,
4396 	.setsockopt =	packet_setsockopt,
4397 	.getsockopt =	packet_getsockopt,
4398 #ifdef CONFIG_COMPAT
4399 	.compat_setsockopt = compat_packet_setsockopt,
4400 #endif
4401 	.sendmsg =	packet_sendmsg,
4402 	.recvmsg =	packet_recvmsg,
4403 	.mmap =		packet_mmap,
4404 	.sendpage =	sock_no_sendpage,
4405 };
4406 
4407 static const struct net_proto_family packet_family_ops = {
4408 	.family =	PF_PACKET,
4409 	.create =	packet_create,
4410 	.owner	=	THIS_MODULE,
4411 };
4412 
4413 static struct notifier_block packet_netdev_notifier = {
4414 	.notifier_call =	packet_notifier,
4415 };
4416 
4417 #ifdef CONFIG_PROC_FS
4418 
4419 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4420 	__acquires(RCU)
4421 {
4422 	struct net *net = seq_file_net(seq);
4423 
4424 	rcu_read_lock();
4425 	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4426 }
4427 
4428 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4429 {
4430 	struct net *net = seq_file_net(seq);
4431 	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4432 }
4433 
4434 static void packet_seq_stop(struct seq_file *seq, void *v)
4435 	__releases(RCU)
4436 {
4437 	rcu_read_unlock();
4438 }
4439 
4440 static int packet_seq_show(struct seq_file *seq, void *v)
4441 {
4442 	if (v == SEQ_START_TOKEN)
4443 		seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4444 	else {
4445 		struct sock *s = sk_entry(v);
4446 		const struct packet_sock *po = pkt_sk(s);
4447 
4448 		seq_printf(seq,
4449 			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4450 			   s,
4451 			   atomic_read(&s->sk_refcnt),
4452 			   s->sk_type,
4453 			   ntohs(po->num),
4454 			   po->ifindex,
4455 			   po->running,
4456 			   atomic_read(&s->sk_rmem_alloc),
4457 			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4458 			   sock_i_ino(s));
4459 	}
4460 
4461 	return 0;
4462 }
4463 
4464 static const struct seq_operations packet_seq_ops = {
4465 	.start	= packet_seq_start,
4466 	.next	= packet_seq_next,
4467 	.stop	= packet_seq_stop,
4468 	.show	= packet_seq_show,
4469 };
4470 
4471 static int packet_seq_open(struct inode *inode, struct file *file)
4472 {
4473 	return seq_open_net(inode, file, &packet_seq_ops,
4474 			    sizeof(struct seq_net_private));
4475 }
4476 
4477 static const struct file_operations packet_seq_fops = {
4478 	.owner		= THIS_MODULE,
4479 	.open		= packet_seq_open,
4480 	.read		= seq_read,
4481 	.llseek		= seq_lseek,
4482 	.release	= seq_release_net,
4483 };
4484 
4485 #endif
4486 
4487 static int __net_init packet_net_init(struct net *net)
4488 {
4489 	mutex_init(&net->packet.sklist_lock);
4490 	INIT_HLIST_HEAD(&net->packet.sklist);
4491 
4492 	if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4493 		return -ENOMEM;
4494 
4495 	return 0;
4496 }
4497 
4498 static void __net_exit packet_net_exit(struct net *net)
4499 {
4500 	remove_proc_entry("packet", net->proc_net);
4501 }
4502 
4503 static struct pernet_operations packet_net_ops = {
4504 	.init = packet_net_init,
4505 	.exit = packet_net_exit,
4506 };
4507 
4508 
4509 static void __exit packet_exit(void)
4510 {
4511 	unregister_netdevice_notifier(&packet_netdev_notifier);
4512 	unregister_pernet_subsys(&packet_net_ops);
4513 	sock_unregister(PF_PACKET);
4514 	proto_unregister(&packet_proto);
4515 }
4516 
4517 static int __init packet_init(void)
4518 {
4519 	int rc = proto_register(&packet_proto, 0);
4520 
4521 	if (rc != 0)
4522 		goto out;
4523 
4524 	sock_register(&packet_family_ops);
4525 	register_pernet_subsys(&packet_net_ops);
4526 	register_netdevice_notifier(&packet_netdev_notifier);
4527 out:
4528 	return rc;
4529 }
4530 
4531 module_init(packet_init);
4532 module_exit(packet_exit);
4533 MODULE_LICENSE("GPL");
4534 MODULE_ALIAS_NETPROTO(PF_PACKET);
4535